Search

Background Whole-body magnetic resonance imaging is accurate, efficient and cost-effective for cancer staging. Machine learning may support radiologists reading whole-body magnetic resonance imaging. Objectives To develop a machine-learning algorithm to detect normal organs and cancer lesions. To compare diagnostic accuracy, time and agreement of radiology reads to detect metastases using whole-body magnetic resonance imaging with concurrent machine learning (whole-body magnetic resonance imaging + machine learning) against standard whole-body magnetic resonance imaging (whole-body magnetic resonance imaging + standard deviation). Design and participants Retrospective analysis of (1) prospective single-centre study in healthy volunteers > 18 years (n = 51) and (2) prospective multicentre STREAMLINE study patient data (n = 438). Tests Index: whole-body magnetic resonance imaging + machine learning. Comparator: whole-body magnetic resonance imaging + standard deviation. Reference standard Previously established expert panel consensus reference at 12 months from diagnosis. Outcome measures Primary: difference in per-patient specificity between whole-body magnetic resonance imaging + machine learning and whole-body magnetic resonance imaging + standard deviation. Secondary: per-patient sensitivity, per-lesion sensitivity and specificity, read time and agreement. Methods Phase 1: classification forests, convolutional neural networks, and a multi-atlas approaches for organ segmentation. Phase 2/3: whole-body magnetic resonance imaging scans were allocated to Phase 2 (training = 226, validation = 45) and Phase 3 (testing = 193). Disease sites were manually labelled. The final algorithm was applied to 193 Phase 3 cases, generating probability heatmaps. Twenty-five radiologists (18 experienced, 7 inexperienced in whole-body magnetic resonance imaging) were randomly allocated whole-body magnetic resonance imaging + machine learning or whole-body magnetic resonance imaging + standard deviation over two or three rounds in a National Health Service setting. Read time was independently recorded. Results Phases 1 and 2: convolutional neural network had best Dice similarity coefficient, recall and precision measurements for healthy organ segmentation. Final algorithm used a ‘two-stage’ initial organ identification followed by lesion detection. Phase 3: evaluable scans (188/193, of which 50 had metastases from 117 colon, 71 lung cancer cases) were read between November 2019 and March 2020. For experienced readers, per-patient specificity for detection of metastases was 86.2% (whole-body magnetic resonance imaging + machine learning) and 87.7% (whole-body magnetic resonance imaging + standard deviation), (difference −1.5%, 95% confidence interval −6.4% to 3.5%; p = 0.387); per-patient sensitivity was 66.0% (whole-body magnetic resonance imaging + machine learning) and 70.0% (whole-body magnetic resonance imaging + standard deviation) (difference −4.0%, 95% confidence interval −13.5% to 5.5%; p = 0.344). For inexperienced readers (53 reads, 15 with metastases), per-patient specificity was 76.3% in both groups with sensitivities of 73.3% (whole-body magnetic resonance imaging + machine learning) and 60.0% (whole-body magnetic resonance imaging + standard deviation). Per-site specificity remained high within all sites; above 95% (experienced) or 90% (inexperienced). Per-site sensitivity was highly variable due to low number of lesions in each site. Reading time lowered under machine learning by 6.2% (95% confidence interval −22.8% to 10.0%). Read time was primarily influenced by read round with round 2 read times reduced by 32% (95% confidence interval 20.8% to 42.8%) overall with subsequent regression analysis showing a significant effect (p = 0.0281) by using machine learning in round 2 estimated as 286 seconds (or 11%) quicker. Interobserver variance for experienced readers suggests moderate agreement, Cohen’s κ = 0.64, 95% confidence interval 0.47 to 0.81 (whole-body magnetic resonance imaging + machine learning) and Cohen’s κ = 0.66, 95% confidence interval 0.47 to 0.81 (whole-body magnetic resonance imaging + standard deviation). Limitations Patient whole-body magnetic resonance imaging data were heterogeneous with relatively few metastatic lesions in a wide variety of locations, making training and testing difficult and hampering evaluation of sensitivity. Conclusions There was no difference in diagnostic accuracy for whole-body magnetic resonance imaging radiology reads with or without machine-learning support, although radiology read time may be slightly shortened using whole-body magnetic resonance imaging + machine learning. Future work Failure-case analysis to improve model training, automate lesion segmentation and transfer of machine-learning techniques to other tumour types and imaging modalities. Study registration This study is registered as ISRCTN23068310. Funding This award was funded by the National Institute for Health and Care Research (NIHR) Efficacy and Mechanism Evaluation (EME) programme (NIHR award ref: 13/122/01) and is published in full in Efficacy and Mechanism Evaluation; Vol. 11, No. 15. See the NIHR Funding and Awards website for further award information. Plain language summary Whole-body magnetic resonance imaging demonstrates the entire body and can detect the spread of tumour, without the burden of ionising radiation. Recently, the STREAMLINE study reported that whole-body magnetic resonance imaging is accurate, efficient and cost-effective for cancer staging. However, whole-body magnetic resonance imaging is complex to report. Machine learning is a type of artificial intelligence whereby a computer learns from being given previous data to undertake a task, using techniques such as classification forests, convolutional neural networks, and multi-atlas approaches. Our aim was to develop a machine-learning method to automatically detect lesions on whole-body magnetic resonance imaging to support radiologists by potentially improving their ability to correctly detect disease and reduce the reading time of whole-body magnetic resonance imaging scans in patients with cancer. Firstly, whole-body magnetic resonance imaging scans from 51 healthy volunteers were used to develop machine-learning methods to automatically detect normal organs. Secondly, machine-learning methods were trained to detect cancer lesions, using 271 whole-body magnetic resonance imaging scans from a previous study. Finally, the refined machine-learning technique was tested in 188 patient scans from a previous study, to see if the technique could improve radiology reporting by increasing accuracy and speed in detecting disease. We designed a system to test the accuracy of radiologists reading whole-body magnetic resonance imaging with or without machine-learning support in a near-real clinical National Health Service setting. Twenty-five independent radiologists (18 experienced in reading whole-body magnetic resonance imaging and 7 radiologists inexperienced in whole-body magnetic resonance imaging) were randomly allocated whole-body magnetic resonance imaging scans to read with or without machine-learning support. We found that machine-learning support resulted in similar accuracy for detecting disease and was slightly more efficient in the reading time than for radiological reads without machine-learning support. Differences in interpretation between experienced readers were considered moderate in both cases. Overall, the study was an ambitious attempt to undertake a highly complex machine-learning task, to detect cancer on whole-body magnetic resonance imaging. Many important steps have been taken but the current machine-learning algorithm did not result in a significant improvement in the radiologist’s accuracy for disease detection, although it may have slightly reduced the time taken to read the study. Future work is advocated to further develop machine-learning tools to improve the accuracy of tumour detection. Scientific summary Background Whole-body magnetic resonance imaging (WB-MRI) has been developed in the last decade and it has been proposed as an alternative to multimodality cancer staging pathways. The STREAMLINE study has demonstrated that staging of lung and colon cancer via WB-MRI was of similar accuracy to standard care staging pathways but resulted in fewer tests being required, and a reduction in staging time and cost. However, this technique has not widely translated into clinical practice, being limited to a few expert centres. A barrier to translation may be the complexity of interpretation by inexperienced readers. A machine-learning (ML) algorithm for automated detection of cancer lesions, to assist radiologists, may allow clinical translation of WB-MRI for the benefit of patient care. Objective The objective of Phases 1 and 2 was to develop a ML method for automatic detection of cancer lesions on WB-MRI. The primary objective of Phase 3 was to compare the diagnostic test accuracy of WB-MRI in patients being staged for cancer with and without ML support, when read by independent, experienced readers. The reference standard was the consensus reference panel from the STREAMLINE study, which had recorded the sites of disease in the STREAMLINE patients using all available clinical information for a 12-month follow-up from time of enrolment. The planned secondary objectives of this study were: to compare the reading time of WB-MRI scans, with and without ML support; to determine the interobserver variability of WB-MRI diagnosis by different radiologists, with and without ML support; to evaluate the diagnostic accuracy of WB-MRI read by non-experienced readers, with and without ML support; to evaluate different combinations of acquired MRI sequences; all the above with and without ML support (not achieved); to determine any difference in costs related to radiology reading time by means of a simple cost-effectiveness analysis (not achieved). Design This was an observational study (study limited to working with data), using different patient cohorts, and methodologies, being evaluated in series during three consecutive phases. Phase 1: Previously acquired WB-MRI scans in 51 healthy volunteers were stitched into imaging volumes. The normal organs and skeleton were segmented by a trained radiologist. We compared and tested several state-of-the-art medical image segmentation methods to train an algorithm to automatically detect and segment the organs. This included a multi-atlas (MA) approach, classification forests (CFs) and convolutional neural networks (CNNs) methods. For the CFs method, we used 50 trees with a maximum tree depth of 30 in the segmentation. The stopping criterion for growing trees was if either the objective function (information gain) could not be further improved or the number of training samples in a leaf fell below a threshold of four samples. For the CNNs method, we employed a dual pathway (2 resolutions), 11-layer deep CNN, where the last 2 layers correspond to fully connected layers, which combine the features extracted on the 2 resolution pathways. We adopted 50–70 feature maps (different kernels) for each layer. The network architecture was fully convolutional and there were no max-pooling layers, which we found to increase segmentation accuracy. The CNN architecture was a balance between model capacity, training efficiency, and memory demands. The third algorithm was based on a MA label propagation approach. MA segmentation uses a set of atlases (images with corresponding segmentations) that represent the interparticipant variability of the anatomy to be segmented. Each atlas was registered to the new image to be segmented using a deformable image registration. The MA approach accounts for anatomical shape variability and is more robust than single-atlas propagation methods, in that at any errors associated with propagation were averaged out when combining multiple atlases. The approach employed here makes use of efficient 3D–3D intensity-based image registration with free-form deformations as the transformation model and correlation coefficient as the similarity measure. Majority voting is used to derive the final tissue label at each voxel. Phase 2: Available WB-MRI scans from the STREAMLINE study (n = 438) were allocated by the study statistician to Phase 2 (model training) and clinical validation (hold-out set of 193 cases). The visible lesions in cases allocated to Phase 2 were segmented by trained radiologists, using the defined consensus reference standard for the sites of disease. Using 226 evaluable and annotated cases, we trained a model for lesion detection and localisation rather than attempting accurate automated segmentation. We found that lesion detection in WB-MRI was suboptimal with the CNN alone. An optimal ‘two-stage’ ML method was developed and tested. In the first stage of the ‘two-stage’ method, the information from Phase 1 was used to identify the position of organs and bones and in stage two, the lesions were detected. Stage two could be modular with respect to the anatomical location in which the suspected lesion could be found. The architecture and configuration of the used CNN were modified to achieve optimal performance. Phase 3: The final two-stage ML algorithm from Phase 2 was applied to the 193 cases that were held out for Phase 3, generating probability heatmap volumes. WB-MRI scans were reported by 25 independent radiologists (18 radiologists experienced in reading WB-MRI and 7 radiologists inexperienced in WB-MRI). All radiologists took part in the 1st and 2nd round reads which incorporated inter-rater reads, and 8 participated in the 3rd round read for intrarater reads. All reads were undertaken in an NHS radiology reporting room, although using a separate cloud-based picture archiving and communication system (PACS). Based on experience level, 10–16 cases were randomly allocated by the study statistician to the reading lists for inexperienced or experienced radiologists. Each radiologist read both lung and colon cases with and without ML as a paired cohort. There were at least 4 weeks interval between each reading round for individual radiologists. Cases allocated to round 1 would be a random mixture of cases with and without ML support and then in round 2 the cases would be reversed such that each case was read once with and once without ML support. The number of cases with and without ML output was balanced in each reading round. A scribe recorded reading time. Results Phase 1 results We found that CNNs outperformed CFs and the MA algorithm when T2w volumes were used as input to the algorithms and when using pooled overlap-evaluation metrics [Dice similarity coefficient (DSC), recall (RE), precision (PR)] to assess the accuracy of segmentation. When the performance of the algorithms was assessed, with pooled surface distance metrics [average surface distance (ASD), root-mean-square surface distance (RMSSD), and Hausdorff distance (HD)], it was the MA algorithm that performed best. Single misinterpreted voxels in CFs and CNNs can greatly elevate ASD, RMSSD, and HD; these metrics are particularly sensitive to outliers. We then assessed the pooled metrics performance of CFs and CNNs when using all imaging combinations [T2w + T1w + diffusion-weighted imaging (DWI)] as input, arguing that maximisation of training information to the algorithms might improve the performance of segmentation. We found that the performance of CFs was improved, however not significantly, when using all imaging combinations as input for training. The opposite was observed for CNNs. The findings for the pooled metrics analysis, described above, were corroborated by a ‘per-organ’ quantitative analysis of the commonly used DSC, to assess the performance of our segmentation algorithms. This analysis confirmed that for all individual anatomical structures (except for the bladder), the algorithm that returned the greatest DSC was CNNs with T2w images only used as input. As our morphological T2w and T1w scans were acquired using breath-holds and the DWI sequence was acquired with free breathing, we found that there was significant displacement between soft tissues in anatomical areas adjacent to the diaphragm between these types of scans. As the employed affine registration method could not fully compensate for nonlinear motions caused by breathing, we assumed that misregistration could be the reason why the performance of CNNs, despite performing better than the other two algorithms when using T2w volumes as input only, was degraded when using all imaging combinations as input for training. A more robust, nonlinear registration method could improve the accuracy of CNNs and further improve the performance of CFs. The performance of our methods cannot be directly compared to similar methods in the literature because there is no previous work describing automatic, simultaneous segmentation of healthy organs and bones in multiparametric WB-MRI. We believe, however, that our methods may compare favourably to other ML methods for detection and segmentation in medical imaging because our classifiers are inherently multilabel and effective training was achieved when using a relatively small number of data sets, something that is very important in the clinical setting. However, we still need to address the performance limitations of our algorithms when segmenting organs with big variability in appearance (e.g. the gallbladder or the pancreas). Phase 2 results We tested different ML methods for lesion detection. We found that using CNNs was not optimal for lesion detection on WB-MRI. This may have been due to the small fraction of lesion volume occupying the scanned space, when compared to the WB volume. It may also have been due to the complexity of intensities in background tissue and the lesion with weak boundaries causing challenges for the CNNs. We, therefore, adapted our process to become an optimal two-stage process, with an initial stage for detection of organs as per the Phase 1 technique followed by a CNN to detect lesions. Stage two could be modular with respect to the anatomical location where the suspected lesion can be found. The architecture and configuration of the used CNN could be modified to achieve optimal performance for lesion detection. Phase 3 results All radiology reads were completed between November 2019 and March 2020. Among the 193 cases allocated to Phase 3, 188 WB-MRI scans were evaluable (117 colon and 71 lung cancer) and 50/188 cases had metastases. Per-patient specificity for detection of metastases within experienced readers was 86.2% (WB-MRI + ML) and 87.7% [WB-MRI + standard deviation (SD)], [difference −1.5%, 95% confidence interval (CI) −6.4%, 3.5%; p = 0.387]. Per-patient sensitivity produced results of 66.0% (WB-MRI + ML) and 70.0% (WB-MRI + SD) (difference −4.0%, 95% CI −13.5% to 5.5%; p = 0.344). For inexperienced readers (53 reads, 15 with metastases), per-patient specificity was 76.3% in both groups with sensitivities of 73.3% (WB-MRI + ML) and 60.0% (WB-MRI + SD). Per-site specificity remained high within all sites; above 95% (experienced) or 90% (inexperienced). Per-site sensitivity was highly variable due to the low number of lesions in each site, hampering interpretation. Reading time was lowered under ML by 6.2% (95% CI −22.8% to 10.0%). Read time was primarily influenced by read round with round 2 read times reduced by 32% (95% CI 20.8% to 42.8%) overall with subsequent regression analysis showing a significant effect (p = 0.0281) by using ML in round 2 estimated as 286 seconds (or 11%) quicker. Interobserver variance for experienced readers suggests moderate agreement, Cohen’s κ = 0.64, 95% CI 0.47 to 0.81 (WB-MRI + ML) and Cohen’s κ = 0.66, 95% CI 0.47 to 0.81 (WB-MRI + SD). Conclusion Phase 1 In Phase 1, we developed and evaluated three state-of-the-art algorithms that automatically segment healthy organs and bones in WB-MRI with accuracy comparable to the one achieved manually by clinical experts, using relatively sparse training data. An algorithm based on CNNs and trained using T2w-only images as input performs favourably when compared to CFs or a MA algorithm, trained with either T2w-only images or a combination of imaging inputs (T2w + T1w + DWI). Using multimodal MRI data as input for training did not improve the segmentation performance in this work, but it is anticipated to improve the segmentation performance if more effective WB registration between the various imaging modalities can be performed. This investigation was the first step towards developing robust algorithms for the automatic detection and segmentation of benign and malignant lesions in WB-MRI scans for staging of cancer patients. Phase 2 There were many challenges in training the algorithm for lesion detection, with a very heterogeneous data set, acquired at 16 sites on different machines. The low number of metastatic lesions were scattered through many anatomic sites. Identification of lesions of relatively small volume in relation to the large volume of the WB-MRI required a two-stage approach for lesion detection, with initial organ identification followed by lesion detection. Phase 3 We undertook a robust diagnostic test accuracy study comparing WB-MRI with ML support (index text) with standard WB-MRI, without ML support (comparator test), with paired reads separated by a wash-out period. Although there was no clear statistical difference with or without ML support in terms of diagnostic test accuracy, either in experienced or inexperienced hands, we found that ML reads were likely to be a little shorter in reading time. Implications for health care Machine learning support for image interpretation is a rapidly expanding area of research. With continually increasing demand for diagnostic imaging and a radiology workforce crisis in the NHS, as well as globally, ML techniques may offer support to allow complex imaging modalities, such as WB-MRI, to be ready for translation into clinical care for the benefit of patients. Phase 1 demonstrated that with relatively sparse data, we could develop very successful organ segmentation tools in highly complex data sets and this has many potential future roles. In this study, we used this as a first step prior to lesion detection and this generic method could be applied to a wider range of disease areas, including other tumour types. We found that ML support slightly shorted the reading time, although with no improvement in diagnostic accuracy. However, we have shown that lesion detection using ML is achievable on heterogeneous and complex multiparametric WB-MRI scans. With further model training, we believe that ML support is feasible in the future, with the potential to improve translation of WB-MRI more widely. In addition, many of the techniques that have been developed in the course of the study have the potential to be applied to other areas of diagnostic imaging. Recommendations for future research Future research should investigate: Further improvement in lesion detection accuracy, with evaluation of failure cases for improved model training. Evaluate lesion detection technique on other cancers, notably breast, prostate and myeloma. Further developments in understanding the marrow composition in healthy aging and in metastatic cancer. Developing further understanding and methods to overcome registration challenges between breath-hold and non-breath-hold MRI sequences in order to improve ML tissue characterisation. Study registration This study is registered as ISRCTN23068310. Funding This award was funded by the National Institute for Health and Care Research (NIHR) Efficacy and Mechanism Evaluation (EME) programme (NIHR award ref: 13/122/01) and is published in full in Efficacy and Mechanism Evaluation; Vol. 11, No. 15. See the NIHR Funding and Awards website for further award information.

Abstract Background The impact of molecular imaging (MI) on patient management after biochemical recurrence (BCR) following radical prostatectomy has been described in many studies. However, it is not known if MI-induced management changes are appropriate. This study aimed to determine if androgen deprivation therapy (ADT) management plan is improved by MI in patients who are candidates for salvage radiation therapy. Methods Data were analyzed from the multicenter prospective PROPS trial evaluating PSMA/Choline PET in patients being considered for salvage radiotherapy (sRT) with BCR after prostatectomy. We compared the pre- and post-MI ADT management plans for each patient and cancer outcomes as predicted by the MSKCC nomogram. A higher percentage of predicted BCR associated with ADT treatment intensification after MI was considered as an improvement in a patient’s management. Results Seventy-three patients with a median PSA of 0.38 ng/mL were included. In bivariate analysis, a positive finding on MI (local or metastatic) was associated with decision to use ADT with an odds ratio of 3.67 (95% CI, 1.25 to 10.71; p = 0.02). No factor included in the nomogram was associated with decision to use ADT. Also, MI improved selection of patients to receive ADT based on predicted BCR after sRT : the predicted nomogram 5-year biochemical-free survivals were 52.5% and 43.3%, (mean difference, 9.2%; 95% CI 0.8 to 17.6; p = 0.03) for sRT alone and ADT±sRT subgroups, while there was no statistically significant difference between subgroups before MI. Conclusions PSMA and/or Choline PET/CT before sRT can potentially improve patient ADT management by directing clinicians towards more appropriate intensification.

Abstract This work presents a biophysical model of diffusion and relaxation MRI for prostate called relaxation vascular, extracellular and restricted diffusion for cytometry in tumours (rVERDICT). The model includes compartment-specific relaxation effects providing T1/T2 estimates and microstructural parameters unbiased by relaxation properties of the tissue. 44 men with suspected prostate cancer (PCa) underwent multiparametric MRI (mp-MRI) and VERDICT-MRI followed by targeted biopsy. We estimate joint diffusion and relaxation prostate tissue parameters with rVERDICT using deep neural networks for fast fitting. We tested the feasibility of rVERDICT estimates for Gleason grade discrimination and compared with classic VERDICT and the apparent diffusion coefficient (ADC) from mp-MRI. The rVERDICT intracellular volume fraction fic discriminated between Gleason 3 + 3 and 3 + 4 (p = 0.003) and Gleason 3 + 4 and ≥ 4 + 3 (p = 0.040), outperforming classic VERDICT and the ADC from mp-MRI. To evaluate the relaxation estimates we compare against independent multi-TE acquisitions, showing that the rVERDICT T2 values are not significantly different from those estimated with the independent multi-TE acquisition (p > 0.05). Also, rVERDICT parameters exhibited high repeatability when rescanning five patients (R2 = 0.79–0.98; CV = 1–7%; ICC = 92–98%). The rVERDICT model allows for accurate, fast and repeatable estimation of diffusion and relaxation properties of PCa sensitive enough to discriminate Gleason grades 3 + 3, 3 + 4 and ≥ 4 + 3.

Objective In men with a raised prostate-specific antigen (PSA), MRI increases the detection of clinically significant cancer and reduces overdiagnosis, with fewer biopsies. MRI as a screening tool has not been assessed independently of PSA in a formal screening study. We report a systematic community-based assessment of the prevalence of prostate MRI lesions in an age-selected population.Methods and analysis Men aged 50–75 were identified from participating general practice (GP) practices and randomly selected for invitation to a screening MRI and PSA. Men with a positive MRI or a raised PSA density (≥0.12 ng/mL2) were recommended for standard National Health Service (NHS) prostate cancer assessment.Results Eight GP practices sent invitations to 2096 men. 457 men (22%) responded and 303 completed both screening tests. Older white men were most likely to respond to the invitation, with black men having 20% of the acceptance rate of white men.One in six men (48/303 men, 16%) had a positive screening MRI, and an additional 1 in 20 men (16/303, 5%) had a raised PSA density alone. After NHS assessment, 29 men (9.6%) were diagnosed with clinically significant cancer and 3 men (1%) with clinically insignificant cancer.Two in three men with a positive MRI, and more than half of men with clinically significant disease had a PSA

Introduction Prostate MRI is a well-established tool for the diagnostic work-up for men with suspected prostate cancer (PCa). Current recommendations advocate the use of multiparametric MRI (mpMRI), which is composed of three sequences: T2-weighted sequence (T2W), diffusion-weighted sequence (DWI) and dynamic contrast-enhanced sequence (DCE). Prior studies suggest that a biparametric MRI (bpMRI) approach, omitting the DCE sequences, may not compromise clinically significant cancer detection, though there are limitations to these studies, and it is not known how this may affect treatment eligibility. A bpMRI approach will reduce scanning time, may be more cost-effective and, at a population level, will allow more men to gain access to an MRI than an mpMRI approach.Methods Prostate Imaging Using MRI±Contrast Enhancement (PRIME) is a prospective, international, multicentre, within-patient diagnostic yield trial assessing whether bpMRI is non-inferior to mpMRI in the diagnosis of clinically significant PCa. Patients will undergo the full mpMRI scan. Radiologists will be blinded to the DCE and will initially report the MRI using only the bpMRI (T2W and DWI) sequences. They will then be unblinded to the DCE sequence and will then re-report the MRI using the mpMRI sequences (T2W, DWI and DCE). Men with suspicious lesions on either bpMRI or mpMRI will undergo prostate biopsy. The main inclusion criteria are men with suspected PCa, with a serum PSA of ≤20 ng/mL and without prior prostate biopsy. The primary outcome is the proportion of men with clinically significant PCa detected (Gleason score ≥3+4 or Gleason grade group ≥2). A sample size of at least 500 patients is required. Key secondary outcomes include the proportion of clinically insignificant PCa detected and treatment decision.Ethics and dissemination Ethical approval was obtained from the National Research Ethics Committee West Midlands, Nottingham (21/WM/0091). Results of this trial will be disseminated through peer-reviewed publications. Participants and relevant patient support groups will be informed about the results of the trial.Trial registration number NCT04571840.

Imaging plays a fundamental role in all aspects of the cancer management pathway. However, conventional imaging techniques are largely reliant on morphological and size descriptors that have well‐known limitations, particularly when considering targeted‐therapy response monitoring. Thus, new imaging methods have been developed to characterise cancer and are now routinely implemented, such as diffusion‐weighted imaging, dynamic contrast enhancement, positron emission technology (PET) and magnetic resonance spectroscopy. However, despite the improvement these techniques have enabled, limitations still remain. Novel imaging methods are now emerging, intent on further interrogating cancers. These techniques are at different stages of maturity along the biomarker pathway and aim to further evaluate the cancer microstructure (vascular, extracellular and restricted diffusion for cytometry in tumours) magnetic resonance imaging (MRI), luminal water fraction imaging] as well as the metabolic alterations associated with cancers (novel PET tracers, hyperpolarised MRI). Finally, the use of machine learning has shown powerful potential applications. By using prostate cancer as an exemplar, this Review aims to showcase these potentially potent imaging techniques and what stage we are at in their application to conventional clinical practice.

Background: Multiparametric magnetic resonance imaging (mpMRI) has improved risk stratification for suspected prostate cancer in patients following prior biopsy. However, not all significant cancers are detected by mpMRI. The PICTURE study provides the ideal opportunity to investigate cancer undetected by mpMRI owing to the use of 5 mm transperineal template mapping (TTPM) biopsy. Objective: To summarise attributes of cancers systematically undetected by mpMRI in patients with prior biopsy. Design, setting, and participants: PICTURE was a paired-cohort confirmatory study in which men requiring repeat biopsy underwent mpMRI followed by TTPM biopsy. Outcome measurements and statistical analysis: Attributes were compared between cancers detected and undetected by mpMRI at the patient level. Four predefined histopathological thresholds were used as the target condition for TTPM biopsy. Application of prostate-specific antigen density (PSAD) was explored. Results and limitations: When nonsuspicious mpMRI was defined as Likert score 1–2, 2.9% of patients (3/103; 95% confidence interval [CI] 0.6–8.3%) with definition 1 disease (Gleason ≥ 4 + 3 of any length or maximum cancer core length [MCCL] ≥ 6 mm of any grade) had their cancer not detected by mpMRI. This proportion was 6.5% (11/168; 95% CI 3.3–11%) for definition 2 disease (Gleason ≥ 3 + 4 of any length or MCCL ≥ 4 mm of any grade), 4.8% (7/146; 95% CI 2.0–9.6%) for any amount of Gleason ≥ 3 + 4 cancer, and 9.3% (20/215; 95% CI 5.8–14%) for any cancer. Definition 1 cancers undetected by mpMRI had lower overall Gleason score (p = 0.02) and maximum Gleason score (p = 0.01) compared to cancers detected by mpMRI. Prostate cancers undetected by mpMRI had shorter MCCL than cancers detected by mpMRI for every cancer threshold: definition 1, 6 versus 8 mm (p = 0.02); definition 2, 5 versus 6 mm (p = 0.04); any Gleason ≥ 3 + 4, 5 versus 6 mm (p = 0.03); and any cancer, 3 versus 5 mm (p = 0.0009). A theoretical PSAD threshold of 0.15 ng/ml/ml reduced the proportion of patients with undetected disease on nonsuspicious mpMRI to 0% (0/105; 95% CI 0–3.5%) for definition 1, 0.58% (1/171; 95% CI 0.01–3.2%) for definition 2, and 0% (0/146) for any Gleason ≥ 3 + 4. Conclusions: Few significant cancers are undetected by mpMRI in patients requiring repeat prostate biopsy. Undetected tumours are of lower overall and maximum Gleason grade and shorter cancer length compared to cancers detected by mpMRI. Patient summary: In patients with a previous prostate biopsy, magnetic resonance imaging (MRI) overlooks few prostate cancers, and these tend to be smaller and less aggressive than cancer that is detected.

Abstract Prostate magnetic resonance imaging (MRI) of high diagnostic quality is a key determinant for either detection or exclusion of prostate cancer. Adequate high spatial resolution on T2-weighted imaging, good diffusion-weighted imaging and dynamic contrast-enhanced sequences of high signal-to-noise ratio are the prerequisite for a high-quality MRI study of the prostate. The Prostate Imaging Quality (PI-QUAL) score was created to assess the diagnostic quality of a scan against a set of objective criteria as per Prostate Imaging-Reporting and Data System recommendations, together with criteria obtained from the image. The PI-QUAL score is a 1-to-5 scale where a score of 1 indicates that all MR sequences (T2-weighted imaging, diffusion-weighted imaging and dynamic contrast-enhanced sequences) are below the minimum standard of diagnostic quality, a score of 3 means that the scan is of sufficient diagnostic quality, and a score of 5 implies that all three sequences are of optimal diagnostic quality. The purpose of this educational review is to provide a practical guide to assess the quality of prostate MRI using PI-QUAL and to familiarise the radiologist and all those involved in prostate MRI with this scoring system. A variety of images are also presented to demonstrate the difference between suboptimal and good prostate MR scans.

Abstract Objectives National guidelines recommend prostate multiparametric (mp) MRI in men with suspected prostate cancer before biopsy. In this study, we explore prostate mpMRI protocols across 14 London hospitals and determine whether standardisation improves diagnostic quality. Methods An MRI physicist facilitated mpMRI set-up across several regional hospitals, working together with experienced uroradiologists who judged diagnostic quality. Radiologists from the 14 hospitals participated in the assessment and optimisation of prostate mpMRI image quality, assessed according to both PiRADSv2 recommendations and on the ability to “rule in” and/or “rule out” prostate cancer. Image quality and sequence parameters of representative mpMRI scans were evaluated across 23 MR scanners. Optimisation visits were performed to improve image quality, and 2 radiologists scored the image quality pre- and post-optimisation. Results 20/23 mpMRI protocols, consisting of 111 sequences, were optimised by modifying their sequence parameters. Pre-optimisation, only 15% of T2W images were non-diagnostic, whereas 40% of ADC maps, 50% of high b-value DWI and 41% of DCE-MRI were considered non-diagnostic. Post-optimisation, the scores were increased with 80% of ADC maps, 74% of high b-value DWI and 88% of DCE-MRI to be partially or fully diagnostic. T2W sequences were not optimised, due to their higher baseline quality scores. Conclusions Targeted intervention at a regional level can improve the diagnostic quality of prostate mpMRI protocols, with implications for improving prostate cancer detection rates and targeted biopsies.

Introduction Multiparametric MRI (mpMRI) is now widely used to risk stratify men with a suspicion of prostate cancer and identify suspicious regions for biopsy. However, the technique has modest specificity and a high false-positive rate, especially in men with mpMRI scored as indeterminate (3/5) or likely (4/5) to have clinically significant cancer (csPCa) (Gleason ≥3+4). Advanced MRI techniques have emerged which seek to improve this characterisation and could predict biopsy results non-invasively. Before these techniques are translated clinically, robust histological and clinical validation is required.Methods and analysis This study aims to clinically validate two advanced MRI techniques in a prospectively recruited cohort of men suspected of prostate cancer. Histological analysis of men undergoing biopsy or prostatectomy will be used for biological validation of biomarkers derived from Vascular and Extracellular Restricted Diffusion for Cytometry in Tumours and Luminal Water imaging. In particular, prostatectomy specimens will be processed using three-dimension printed patient-specific moulds to allow for accurate MRI and histology mapping. The index tests will be compared with the histological reference standard to derive false positive rate and true positive rate for men with mpMRI scores which are indeterminate (3/5) or likely (4/5) to have clinically significant prostate cancer (csPCa). Histopathological validation from both biopsy and prostatectomy samples will provide the best ground truth in validating promising MRI techniques which could predict biopsy results and help avoid unnecessary biopsies in men suspected of prostate cancer.Ethics and dissemination Ethical approval was granted by the London—Queen Square Research Ethics Committee (19/LO/1803) on 23 January 2020. Results from the study will be presented at conferences and submitted to peer-reviewed journals for publication. Results will also be available on ClinicalTrials.gov.Trial registration number NCT04792138.

Abstract Medical therapies achieve their control at expense to the patient in the form of a range of toxicities, which incur costs and diminish quality of life. Magnetic resonance navigation is an emergent technique that enables image‐guided remote‐control of magnetically labeled therapies and devices in the body, using a magnetic resonance imaging (MRI) system. Minimally INvasive IMage‐guided Ablation (MINIMA), a novel, minimally invasive, MRI‐guided ablation technique, which has the potential to avoid traditional toxicities, is presented. It comprises a thermoseed navigated to a target site using magnetic propulsion gradients generated by an MRI scanner, before inducing localized cell death using an MR‐compatible thermoablative device. The authors demonstrate precise thermoseed imaging and navigation through brain tissue using an MRI system (0.3 mm), and they perform thermoablation in vitro and in vivo within subcutaneous tumors, with the focal ablation volume finely controlled by heating duration. MINIMA is a novel theranostic platform, combining imaging, navigation, and heating to deliver diagnosis and therapy in a single device.

Introduction The ReIMAGINE Consortium was conceived to develop risk-stratification models that might incorporate the full range of novel prostate cancer (PCa) diagnostics (both commercial and academic). Methods ReIMAGINE Risk is an ethics approved (19/LO/1128) multicentre, prospective, observational cohort study which will recruit 1000 treatment-naive men undergoing a multi-parametric MRI (mpMRI) due to an elevated PSA (≤20ng/ml) or abnormal prostate examination who subsequently had a suspicious mpMRI (score≥3, stage ≤T3bN0M0). Primary outcomes include the detection of ≥Gleason 7 PCa at baseline and time to clinical progression, metastasis and death. Baseline blood, urine, and biopsy cores for fresh prostate tissue samples (2 targeted and 1 non-targeted) will be biobanked for future analysis. High-resolution scanning of pathology whole-slide imaging and MRI-DICOM images will be collected. Consortium partners will be granted access to data and biobanks to develop and validate biomarkers using correlation to mpMRI, biopsy-based disease status and long-term clinical outcomes. Results Recruitment began in September 2019(n = 533). A first site opened in September 2019 (n = 296), a second in November 2019 (n = 210) and a third in December 2020 (n = 27). Acceptance to the study has been 65% and a mean of 36.5ml(SD+/-10.0), 12.9ml(SD+/-3.7) and 2.8ml(SD+/-0.7) urine, plasma and serum donated for research, respectively. There are currently 4 academic and 15 commercial partners spanning imaging (~9 radiomics, artificial intelligence/machine learning), fluidic (~3 blood-based and ~2urine-based) and tissue-based (~1) biomarkers. Conclusion The consortium will develop, or adjust, risk models for PCa, and provide a platform for evaluating the role of novel diagnostics in the era of pre-biopsy MRI and targeted biopsy.

Purpose: We investigate the feasibility of data-driven, model-free quantitative MRI (qMRI) protocol design on in vivo brain and prostate diffusion-relaxation imaging (DRI).Methods: We select subsets of measurements within lengthy pilot scans, without identifying tissue parameters for which to optimise for. We use the “select and retrieve via direct upsampling” (SARDU-Net) algorithm, made of a selector, identifying measurement subsets, and a predictor, estimating fully-sampled signals from the subsets. We implement both using artificial neural networks, which are trained jointly end-to-end. We deploy the algorithm on brain (32 diffusion-/T1-weightings) and prostate (16 diffusion-/T2-weightings) DRI scans acquired on three healthy volunteers on two separate 3T Philips systems each. We used SARDU-Net to identify sub-protocols of fixed size, assessing reproducibility and testing sub-protocols for their potential to inform multi-contrast analyses via the T1-weighted spherical mean diffusion tensor (T1-SMDT, brain) and hybrid multi-dimensional MRI (HM-MRI, prostate) models, for which sub-protocol selection was not optimised explicitly.Results: In both brain and prostate, SARDU-Net identifies sub-protocols that maximise information content in a reproducible manner across training instantiations using a small number of pilot scans. The sub-protocols support T1-SMDT and HM-MRI multi-contrast modelling for which they were not optimised explicitly, providing signal quality-of-fit in the top 5% against extensive sub-protocol comparisons.Conclusions: Identifying economical but informative qMRI protocols from subsets of rich pilot scans is feasible and potentially useful in acquisition-time-sensitive applications in which there is not a qMRI model of choice. SARDU-Net is demonstrated to be a robust algorithm for data-driven, model-free protocol design.

Introduction. Prostate cancer (PCa) is the most frequent cancer diagnosis in men worldwide. Our ability to identify those men whose cancer will decrease their lifespan and/or quality of life remains poor. The ReIMAGINE Consortium has been established to improve PCa diagnosis.Materials and methods. MRI will likely become the future cornerstone of the risk-stratification process for men at risk of early prostate cancer. We will, for the first time, be able to combine the underlying molecular changes in PCa with the state-of-the-art imaging. ReIMAGINE Screening invites men for MRI and PSA evaluation. ReIMAGINE Risk includes men at risk of prostate cancer based on MRI, and includes biomarker testing.Results. Baseline clinical information, genomics, blood, urine, fresh prostate tissue samples, digital pathology and radiomics data will be analysed. Data will be de-identified, stored with correlated mpMRI disease endotypes and linked with long term follow-up outcomes in an instance of the Philips Clinical Data Lake, consisting of cloud-based software. The ReIMAGINE platform includes application programming interfaces and a user interface that allows users to browse data, select cohorts, manage users and access rights, query data, and more. Connection to analytics tools such as Python allows statistical and stratification method pipelines to run profiling regression analyses. Discussion. The ReIMAGINE Multimodal Warehouse comprises a unique data source for PCa research, to improve risk stratification for PCa and inform clinical practice. The de-identified dataset characterized by clinical, imaging, genomics and digital pathology PCa patient phenotypes will be a valuable resource for the scientific and medical community.

Introduction The primary objective of the ReIMAGINE Prostate Cancer Screening Study is to explore the uptake of an invitation to prostate cancer screening using MRI.Methods and analysis The ReIMAGINE Prostate Cancer Screening Study is a prospective single-centre feasibility study. Eligible men aged 50–75 years with no prior prostate cancer diagnosis or treatment will be identified through general practitioner practices and randomly selected for invitation. Those invited will be offered an MRI scan and a prostate-specific antigen (PSA) blood test. The screening MRI scan consists of T2-weighted, diffusion-weighted and research-specific sequences, without the use of intravenous contrast agents. Men who screen positive on either MRI or PSA density will be recommended to have standard of care (National Health Service) tests for prostate cancer assessment, which includes multiparametric MRI. The study will assess the acceptability of an MRI-based prostate screening assessment and the prevalence of cancer detected in MRI-screened men. Summary statistics will be used to explore baseline characteristics in relation to acceptance rates and prevalence of cancer.Ethics and dissemination ReIMAGINE Prostate Cancer Screening is a single-site screening study to assess the feasibility of MRI as a screening tool for prostate cancer. Ethical approval was granted by London–Stanmore Research Ethics Committee Heath Research Authority (reference 19/LO/1129). Study results will be published in peer-reviewed journals after completion of data analysis and used to inform the design of a multicentre screening study in the UK.Trial registration number ClinicalTrials.gov Registry (NCT04063566).

Abstract Background Accurate whole-body staging following biochemical relapse in prostate cancer is vital in determining the optimum disease management. Current imaging guidelines recommend various imaging platforms such as computed tomography (CT), Technetium 99 m (99mTc) bone scan and 18F-choline and recently 68Ga-PSMA positron emission tomography (PET) for the evaluation of the extent of disease. Such approach requires multiple hospital attendances and can be time and resource intensive. Recently, whole-body magnetic resonance imaging (WB-MRI) has been used in a single visit scanning session for several malignancies, including prostate cancer, with promising results, providing similar accuracy compared to the combined conventional imaging techniques. The LOCATE trial aims to investigate the application of WB-MRI for re-staging of patients with biochemical relapse (BCR) following external beam radiotherapy and brachytherapy in patients with prostate cancer. Methods/design The LOCATE trial is a prospective cohort, multi-centre, non-randomised, diagnostic accuracy study comparing WB-MRI and conventional imaging. Eligible patients will undergo WB-MRI in addition to conventional imaging investigations at the time of BCR and will be asked to attend a second WB-MRI exam, 12-months following the initial scan. WB-MRI results will be compared to an enhanced reference standard comprising all the initial, follow-up imaging and non-imaging investigations. The diagnostic performance (sensitivity and specificity analysis) of WB-MRI for re-staging of BCR will be investigated against the enhanced reference standard on a per-patient basis. An economic analysis of WB-MRI compared to conventional imaging pathways will be performed to inform the cost-effectiveness of the WB-MRI imaging pathway. Additionally, an exploratory sub-study will be performed on blood samples and exosome-derived human epidermal growth factor receptor (HER) dimer measurements will be taken to investigate its significance in this cohort. Discussion The LOCATE trial will compare WB-MRI versus the conventional imaging pathway including its cost-effectiveness, therefore informing the most accurate and efficient imaging pathway. Trial registration LOCATE trial was registered on ClinicalTrial.gov on 18th of October 2016 with registration reference number NCT02935816.

False positives on multiparametric MRIs (mp-MRIs) result in many unnecessary invasive biopsies in men with clinically insignificant diseases. This study investigated whether quantitative diffusion MRI could differentiate between false positives, true positives and normal tissue non-invasively. Thirty-eight patients underwent mp-MRI and Vascular, Extracellular and Restricted Diffusion for Cytometry in Tumors (VERDICT) MRI, followed by transperineal biopsy. The patients were categorized into two groups following biopsy: (1) significant cancer—true positive, 19 patients; (2) atrophy/inflammation/high-grade prostatic intraepithelial neoplasia (PIN)—false positive, 19 patients. The clinical apparent diffusion coefficient (ADC) values were obtained, and the intravoxel incoherent motion (IVIM), diffusion kurtosis imaging (DKI) and VERDICT models were fitted via deep learning. Significant differences (p < 0.05) between true positive and false positive lesions were found in ADC, IVIM perfusion fraction (f) and diffusivity (D), DKI diffusivity (DK) (p < 0.0001) and kurtosis (K) and VERDICT intracellular volume fraction (fIC), extracellular–extravascular volume fraction (fEES) and diffusivity (dEES) values. Significant differences between false positives and normal tissue were found for the VERDICT fIC (p = 0.004) and IVIM D. These results demonstrate that model-based diffusion MRI could reduce unnecessary biopsies occurring due to false positive prostate lesions and shows promising sensitivity to benign diseases.

Background: The accuracy of multi-parametric MRI (mpMRI) in the pre-operative staging of prostate cancer (PCa) remains controversial. Objective: The purpose of this study was to evaluate the ability of mpMRI to accurately predict PCa extra-prostatic extension (EPE) on a side-specific basis using a risk-stratified 5-point Likert scale. This study also aimed to assess the influence of mpMRI scan quality on diagnostic accuracy. Patients and Methods: We included 124 men who underwent robot-assisted RP (RARP) as part of the NeuroSAFE PROOF study at our centre. Three radiologists retrospectively reviewed mpMRI blinded to RP pathology and assigned a Likert score (1–5) for EPE on each side of the prostate. Each scan was also ascribed a Prostate Imaging Quality (PI-QUAL) score for assessing the quality of the mpMRI scan, where 1 represents the poorest and 5 represents the best diagnostic quality. Outcome measurements and statistical analyses: Diagnostic performance is presented for the binary classification of EPE, including 95% confidence intervals and the area under the receiver operating characteristic curve (AUC). Results: A total of 231 lobes from 121 men (mean age 56.9 years) were evaluated. Of these, 39 men (32.2%), or 43 lobes (18.6%), had EPE. A Likert score ≥3 had a sensitivity (SE), specificity (SP), NPV, and PPV of 90.4%, 52.3%, 96%, and 29.9%, respectively, and the AUC was 0.82 (95% CI: 0.77–0.86). The AUC was 0.76 (95% CI: 0.64–0.88), 0.78 (0.72–0.84), and 0.92 (0.88–0.96) for biparametric scans, PI-QUAL 1–3, and PI-QUAL 4–5 scans, respectively. Conclusions: MRI can be used effectively by genitourinary radiologists to rule out EPE and help inform surgical planning for men undergoing RARP. EPE prediction was more reliable when the MRI scan was (a) multi-parametric and (b) of a higher image quality according to the PI-QUAL scoring system.

OBJECTIVE:To identify the whole-body MRI (WB-MRI) image type(s) with the highest value for assessment of multiple myeloma, in order to optimise acquisition protocols and read times. METHODS:Thirty patients with clinically-suspected MM underwent WB-MRI at 3 Tesla. Unenhanced Dixon images [fat-only (FO) and water-only (WO)], post contrast Dixon [fat-only plus contrast (FOC) and water-only plus contrast (WOC)] and diffusion weighted images (DWI) of the pelvis from all 30 patients were randomised and read by three experienced readers. For each image type, each reader identified and labelled all visible myeloma lesions. Each identified lesion was compared with a composite reference standard achieved by review of a complete imaging dataset by a further experienced consultant radiologist to determine truly positive lesions. Lesion count, true positives, sensitivity, and positive predictive value were determined. Time to read each scan set was recorded. Confidence for a diagnosis of myeloma was scored using a Likert scale. Conspicuity of focal lesions was assessed in terms of percent contrast and contrast to noise ratio (CNR). RESULTS:Lesion count, true positives, sensitivity and confidence scores were significantly higher when compared to other image types for DWI (P

(1) Background: There is currently limited evidence on the diagnostic accuracy of abbreviated biparametric MRI (a-bpMRI) protocols for prostate cancer (PCa) detection and screening. In the present study, we aim to investigate the performance of a-bpMRI among multiple readers and its potential application to an imaging-based screening setting. (2) Methods: A total of 151 men who underwent 3T multiparametric MRI (mpMRI) of the prostate and transperineal template prostate mapping biopsies were retrospectively selected. Corresponding bpMRI (multiplanar T2WI, DWI, ADC maps) and a-bpMRI (axial T2WI and b 2000 s/mm2 DWI only) dataset were derived from mpMRI. Three experienced radiologists scored a-bpMRI, standard biparametric MRI (bpMRI) and mpMRI in separate sessions. Diagnostic accuracy and interreader agreement of a-bpMRI was tested for different positivity thresholds and compared to bpMRI and mpMRI. Predictive values of a-bpMRI were computed for lower levels of PCa prevalence to simulate a screening setting. The primary definition of clinically significant PCa (csPCa) was Gleason ≥ 4 + 3, or cancer core length ≥ 6 mm. (3) Results: The median age was 62 years, the median PSA was 6.8 ng/mL, and the csPCa prevalence was 40%. Using a cut off of MRI score ≥ 3, the sensitivity and specificity of a-bpMRI were 92% and 48%, respectively. There was no significant difference in sensitivity compared to bpMRI and mpMRI. Interreader agreement of a-bpMRI was moderate (AC1 0.58). For a low prevalence of csPCa (e.g.,

Background: Whole-body magnetic resonance imaging is advocated as an alternative to standard pathways for staging cancer. Objectives: The objectives were to compare diagnostic accuracy, efficiency, patient acceptability, observer variability and cost-effectiveness of whole-body magnetic resonance imaging and standard pathways in staging newly diagnosed non-small-cell lung cancer (Streamline L) and colorectal cancer (Streamline C). Design: The design was a prospective multicentre cohort study. Setting: The setting was 16 NHS hospitals. Participants: Consecutive patients aged ≥ 18 years with histologically proven or suspected colorectal (Streamline C) or non-small-cell lung cancer (Streamline L). Interventions: Whole-body magnetic resonance imaging. Standard staging investigations (e.g. computed tomography and positron emission tomography–computed tomography). Reference standard: Consensus panel decision using 12-month follow-up data. Main outcome measures: The primary outcome was per-patient sensitivity difference between whole-body magnetic resonance imaging and standard staging pathways for metastasis. Secondary outcomes included differences in specificity, the nature of the first major treatment decision, time and number of tests to complete staging, patient experience and cost-effectiveness. Results: Streamline C – 299 participants were included. Per-patient sensitivity for metastatic disease was 67% (95% confidence interval 56% to 78%) and 63% (95% confidence interval 51% to 74%) for whole-body magnetic resonance imaging and standard pathways, respectively, a difference in sensitivity of 4% (95% confidence interval –5% to 13%; p = 0.51). Specificity was 95% (95% confidence interval 92% to 97%) and 93% (95% confidence interval 90% to 96%) respectively, a difference of 2% (95% confidence interval –2% to 6%). Pathway treatment decisions agreed with the multidisciplinary team treatment decision in 96% and 95% of cases, respectively, a difference of 1% (95% confidence interval –2% to 4%). Time for staging was 8 days (95% confidence interval 6 to 9 days) and 13 days (95% confidence interval 11 to 15 days) for whole-body magnetic resonance imaging and standard pathways, respectively, a difference of 5 days (95% confidence interval 3 to 7 days). The whole-body magnetic resonance imaging pathway was cheaper than the standard staging pathway: £216 (95% confidence interval £211 to £221) versus £285 (95% confidence interval £260 to £310). Streamline L – 187 participants were included. Per-patient sensitivity for metastatic disease was 50% (95% confidence interval 37% to 63%) and 54% (95% confidence interval 41% to 67%) for whole-body magnetic resonance imaging and standard pathways, respectively, a difference in sensitivity of 4% (95% confidence interval –7% to 15%; p = 0.73). Specificity was 93% (95% confidence interval 88% to 96%) and 95% (95% confidence interval 91% to 98%), respectively, a difference of 2% (95% confidence interval –2% to 7%). Pathway treatment decisions agreed with the multidisciplinary team treatment decision in 98% and 99% of cases, respectively, a difference of 1% (95% confidence interval –2% to 4%). Time for staging was 13 days (95% confidence interval 12 to 14 days) and 19 days (95% confidence interval 17 to 21 days) for whole-body magnetic resonance imaging and standard pathways, respectively, a difference of 6 days (95% confidence interval 4 to 8 days). The whole-body magnetic resonance imaging pathway was cheaper than the standard staging pathway: £317 (95% confidence interval £273 to £361) versus £620 (95% confidence interval £574 to £666). Participants generally found whole-body magnetic resonance imaging more burdensome than standard imaging but most participants preferred the whole-body magnetic resonance imaging staging pathway if it reduced time to staging and/or number of tests. Limitations: Whole-body magnetic resonance imaging was interpreted by practitioners blinded to other clinical data, which may not fully reflect how it is used in clinical practice. Conclusions: In colorectal and non-small-cell lung cancer, the whole-body magnetic resonance imaging staging pathway has similar accuracy to standard staging pathways, is generally preferred by patients, improves staging efficiency and has lower staging costs. Future work should address the utility of whole-body magnetic resonance imaging for treatment response assessment. Trial registration: Current Controlled Trials ISRCTN43958015 and ISRCTN50436483. Funding: This project was funded by the NIHR Health Technology Assessment programme and will be published in full in Health Technology Assessment; Vol. 23, No. 66. See the NIHR Journals Library website for further project information.

Abstract Background and aims Rapid and accurate cancer staging following diagnosis underpins patient management, in particular the identification of distant metastatic disease. Current staging guidelines recommend sequential deployment of various imaging platforms such as computerised tomography (CT) and positron emission tomography (PET) which can be time and resource intensive and onerous for patients. Recent studies demonstrate that whole body magnetic resonance Imaging (WB-MRI) may stage cancer efficiently in a single visit, with potentially greater accuracy than current staging investigations. The Streamline trials aim to evaluate whether WB-MRI increases per patient detection of metastases in non-small cell lung and colorectal cancer compared to standard staging pathways. Methods The Streamline trials are multicentre, non-randomised, single-arm, prospective diagnostic accuracy studies with a novel design to capture patient management decisions during staging pathways. The two trials recruit adult patients with proven or highly suspected new diagnosis of primary colorectal (Streamline C) or non-small cell lung cancer (Streamline L) referred for staging. Patients undergo WB-MRI in addition to standard staging investigations. Strict blinding protocols are enforced for those interpreting the imaging. A first major treatment decision is made by the multi-disciplinary team prior to WB-MRI revelation based on standard staging investigations only, then based on the WB-MRI and any additional tests precipitated by WB-MRI, and finally based on all available test results. The reference standard is derived by a multidisciplinary consensus panel who assess 12 months of follow-up data to adjudicate on the TNM stage at diagnosis. Health psychology assessment of patients’ experiences of the cancer staging pathway will be undertaken via interviews and questionnaires. A cost (effectiveness) analysis of WB-MRI compared to standard staging pathways will be performed. Discussion We describe a novel approach to radiologist and clinician blinding to ascertain the ‘true’ diagnostic accuracy of differing imaging pathways and discuss our approach to assessing the impact of WB-MRI on clinical decision making in real-time. The Streamline trials will compare WB-MRI and standard imaging pathways in the same patients, thereby informing the most accurate and efficient approach to staging. Trial registration Streamline C ISRCTN43958015 (registered 25/7/2012). Streamline L ISRCTN50436483 (registered 31/7/2012).

BackgroundRisk models (RM) need external validation to assess their value beyond the setting in which they were developed. We validated a RM combining mpMRI and clinical parameters for the probability of harboring significant prostate cancer (sPC, Gleason Score ≥ 3+4) for biopsy-naïve men.Material and methodsThe original RM was based on data of 670 biopsy-naïve men from Heidelberg University Hospital who underwent mpMRI with PI-RADS scoring prior to MRI/TRUS-fusion biopsy 2012-2015. Validity was tested by a consecutive cohort of biopsy-naïve men from Heidelberg (n = 160) and externally by a cohort of 133 men from University College London Hospital (UCLH). Assessment of validity was performed at fusion-biopsy by calibration plots, receiver operating characteristics curve and decision curve analyses. The RM`s performance was compared to ERSPC-RC3, ERSPC-RC3+PI-RADSv1.0 and PI-RADSv1.0 alone.ResultsSPC was detected in 76 men (48%) at Heidelberg and 38 men (29%) at UCLH. The areas under the curve (AUC) were 0.86 for the RM in both cohorts. For ERSPC-RC3+PI-RADSv1.0 the AUC was 0.84 in Heidelberg and 0.82 at UCLH, for ERSPC-RC3 0.76 at Heidelberg and 0.77 at UCLH and for PI-RADSv1.0 0.79 in Heidelberg and 0.82 at UCLH. Calibration curves suggest that prevalence of sPC needs to be adjusted to local circumstances, as the RM overestimated the risk of harboring sPC in the UCLH cohort. After prevalence-adjustment with respect to the prevalence underlying ERSPC-RC3 to ensure a generalizable comparison, not only between the Heidelberg and die UCLH subgroup, the RM`s Net benefit was superior over the ERSPC`s and the mpMRI`s for threshold probabilities above 0.1 in both cohorts.ConclusionsThe RM discriminated well between men with and without sPC at initial MRI-targeted biopsy but overestimated the sPC-risk at UCLH. Taking prevalence into account, the model demonstrated benefit compared with clinical risk calculators and PI-RADSv1.0 in making the decision to biopsy men at suspicion of PC. However, prevalence differences must be taken into account when using or validating the presented risk model.

To determine which of four Dixon image types [in-phase (IP), out-of-phase (OP), fat only (FO) and water-only (WO)] is most sensitive for detecting multiple myeloma (MM) focal lesions on whole body MRI (WB-MRI) images.Thirty patients with clinically-suspected MM underwent WB-MRI at 3 Tesla. Unenhanced IP, OP, FO and WO Dixon images were generated and read by four radiologists. On each image type, each radiologist identified and labelled all visible myeloma lesions in the bony pelvis. Each identified lesion was compared with a reference standard consisting of pre- and post-contrast Dixon and diffusion weighted imaging (read by a further consultant radiologist) to determine whether the lesion was truly positive. Lesion count, true positives, sensitivity, and positive predictive value were compared across the four Dixon image types.Lesion count, true positives, sensitivity and confidence scores were all significantly higher on FO images than on IP images (p>0.05).FO images are more sensitive than other Dixon image types for MM focal lesions, and should be preferentially read by radiologists to improve diagnostic accuracy and reporting efficiency.

MRI has recently been applied as a tool to quantitatively evaluate the response to therapy in patients with Crohn's disease, and is the preferred choice for repeated imaging. Bowel wall thickness on MRI is an important biomarker of underlying inflammatory activity, being abnormally increased in the acute phase and reducing in response to successful therapy; however, a poor level of interobserver agreement of measured thickness is reported and therefore a system for accurate, robust and reproducible measurements is desirable. We propose a novel method for estimating bowel wall-thickness to improve the poor interobserver agreement of the manual procedure. We show that the variability of wall thickness measurement between the algorithm and observer measurements (0.25mm ± 0.81mm) has differences which are similar to observer variability (0.16mm ± 0.64mm).