Search

Supplemental methods

Supplemental Tables S1-S12

Figures S1-S10

Radiomics refers to the high-throughput extraction of quantitative features from radiological scans and is widely used to search for imaging biomarkers for the prediction of clinical outcomes. Current radiomic signatures suffer from limited reproducibility and generalizability, because most features are dependent on imaging modality and tumour histology, making them sensitive to variations in scan protocol. Here, we propose novel radiological features that are specially designed to ensure compatibility across diverse tissues and imaging contrast. These features provide systematic characterization of tumour morphology and spatial heterogeneity. In an international multi-institution study of 1,682 patients, we discover and validate four unifying imaging subtypes across three malignancies and two major imaging modalities. These tumour subtypes demonstrate distinct molecular characteristics and prognoses after conventional therapies. In advanced lung cancer treated with immunotherapy, one subtype is associated with improved survival and increased tumour-infiltrating lymphocytes compared with the others. Deep learning enables automatic tumour segmentation and reproducible subtype identification, which can facilitate practical implementation. The unifying radiological tumour classification may inform prognosis and treatment response for precision medicine. Radiomics has been used to discover imaging signatures that predict therapy response and outcomes, but clinical translation has been slow. Using machine learning methods, the authors report tumour subtypes that are applicable across major imaging modalities and three cancer types. The tumour subtypes have distinct radiological and molecular features, as well as survival outcomes after conventional therapies.

Objective Being able to predict a patient’s life expectancy can help doctors and patients prioritize treatments and supportive care. For predicting life expectancy, physicians have been shown to outperform traditional models that use only a few predictor variables. It is possible that a machine learning model that uses many predictor variables and diverse data sources from the electronic medical record can improve on physicians’ performance. For patients with metastatic cancer, we compared accuracy of life expectancy predictions by the treating physician, a machine learning model, and a traditional model. Materials and Methods A machine learning model was trained using 14 600 metastatic cancer patients’ data to predict each patient’s distribution of survival time. Data sources included note text, laboratory values, and vital signs. From 2015–2016, 899 patients receiving radiotherapy for metastatic cancer were enrolled in a study in which their radiation oncologist estimated life expectancy. Survival predictions were also made by the machine learning model and a traditional model using only performance status. Performance was assessed with area under the curve for 1-year survival and calibration plots. Results The radiotherapy study included 1190 treatment courses in 899 patients. A total of 879 treatment courses in 685 patients were included in this analysis. Median overall survival was 11.7 months. Physicians, machine learning model, and traditional model had area under the curve for 1-year survival of 0.72 (95% CI 0.63–0.81), 0.77 (0.73–0.81), and 0.68 (0.65–0.71), respectively. Conclusions The machine learning model’s predictions were more accurate than those of the treating physician or a traditional model.

The incidence of oropharyngeal squamous cell carcinoma (OPSCC) has been rapidly increasing. Disease stage and smoking history are often used in current clinical trials to select patients for deintensification therapy, but these features lack sufficient accuracy for predicting disease relapse. Our purpose was to develop an imaging signature to assess early response and predict outcomes of OPSCC. Methods: We retrospectively analyzed 162 OPSCC patients treated with concurrent chemoradiotherapy, equally divided into separate training and validation cohorts with similar clinical characteristics. A robust consensus clustering approach was used to spatially partition the primary tumor and involved lymph nodes into subregions (i.e., habitats) based on (18)F-FDG PET and contrast CT imaging. We proposed quantitative image features to characterize the temporal volumetric change of the habitats and peritumoral/nodal tissue between baseline and midtreatment. The reproducibility of these features was evaluated. We developed an imaging signature to predict progression-free survival (PFS) by fitting an L1-regularized Cox regression model. Results: We identified 3 phenotypically distinct intratumoral habitats: metabolically active and heterogeneous, enhancing and heterogeneous, and metabolically inactive and homogeneous. The final Cox model consisted of 4 habitat evolution-based features. In both cohorts, this imaging signature significantly outperformed traditional imaging metrics, including midtreatment metabolic tumor volume for predicting PFS, with a C-index of 0.72 versus 0.67 (training) and 0.66 versus 0.56 (validation). The imaging signature stratified patients into high-risk versus low-risk groups with 2-y PFS rates of 59.1% versus 89.4% (hazard ratio, 4.4; 95% confidence interval, 1.4–13.4 [training]) and 61.4% versus 87.8% (hazard ratio, 4.6; 95% confidence interval, 1.7–12.1 [validation]). The imaging signature remained an independent predictor of PFS in multivariable analysis adjusting for stage, human papillomavirus status, and smoking history. Conclusion: The proposed imaging signature allows more accurate prediction of disease progression and, if prospectively validated, may refine OPSCC patient selection for risk-adaptive therapy.

Randomized clinical trials (RCT) are the gold standard for informing treatment decisions. Observational studies are often plagued by selection bias, and expert-selected covariates may insufficiently adjust for confounding. We explore how unstructured clinical text can be used to reduce selection bias and improve medical practice. We develop a framework based on natural language processing to uncover interpretable potential confounders from text. We validate our method by comparing the estimated hazard ratio (HR) with and without the confounders against established RCTs. We apply our method to four cohorts built from localized prostate and lung cancer datasets from the Stanford Cancer Institute and show that our method shifts the HR estimate towards the RCT results. The uncovered terms can also be interpreted by oncologists for clinical insights. We present this proof-of-concept study to enable more credible causal inference using observational data, uncover meaningful insights from clinical text, and inform high-stakes medical decisions.

PURPOSE/OBJECTIVE(S) The first clinical biology-guided radiation therapy (BgRT) system - RefleXion X1 - is installed and is being commissioned for clinical use at our institution. The system delivers 6MV-FFF beam via 50 firing positions at 850 cGy/min dose rate on 60 rpm rotating gantry with the couch advancing in 2.1 mm increments. Collimation is achieved through a binary multi-leaf collimator system composed of 64 (6.25 mm width at 850 cm SAD) leaves and 10mm or 20mm jaws. The purpose of this study was to compare the treatment plan quality and delivery efficiency for SBRT cases without PET-guidance. MATERIALS/METHODS Five lung cancer patients (50 Gy/4fx) with motion-inclusive PTV previously treated with VMAT SBRT were selected for this retrospective study. For each VMAT plan a corresponding plan was generated on the RefleXion treatment planning system (TPS) using our institutional planning constraints. VMAT SBRT plans used 1-2 partial arcs with 10MV-FFF energy and 2400 MU/min dose rate. RefleXion SBRT plans used all 50 firing positions and 1 cm jaw collimation. As plan normalization option does not yet exist in the RefleXion TPS, the plans were optimized to achieve PTV D95% coverage with 100% of prescription dose to the best of planner's ability. All clinically relevant metrics in this study, including plan PTV D95%, PTV D1%, 50% isodose volume (I50%), Conformity Index (CI), organs at risk constraints, and treatment time were analyzed and compared between VMAT and RefleXion plans using paired t-tests. RESULTS Clinically acceptable plans were obtained with both techniques. The average PTV volume was 14.1 ± 6.4cc. For VMAT and RefleXion plans, no statistically significant difference was observed in PTV D95%, PTV D1%, CI or I50%. Ipsilateral lung volume spared at 12.4 Gy (VS12.4) was slightly greater (31.6cc increase, P < 0.035) and skin Dmax was slightly lower (1.4 Gy decrease, P < 0.038) for RefleXion plans. Treatment beam-on time was significantly longer for RefleXion plans (23.4min increase, P < 0.0002). CONCLUSION RefleXion TPS provided comparable plan quality to lung SBRT VMAT plans. Treatment beam-on time is significantly longer for the RefleXion system.

Purpose/objective(s) The X1 system ("X1") integrates a fully functional Positron Emission Tomography (PET) combined with a fan-beam kVCT system into the architecture of a ring-gantry linear accelerator (LINAC) for biology-guided radiotherapy (BgRT) delivery. The integration of PET into a LINAC framework required design features that deviate from the diagnostic setting. For instance, the X1 houses two opposed 90° PET arcs that rotate at 60 RPM to generate sinogram reconstructed images rather than a static, circumferential array of detectors. To better understand X1 PET performance, we conducted a side-by-side comparison of images of a PET phantom obtained sequentially on the X1 PET subsystem and a traditional PET/CT scanner. Materials/methods An independent quality assurance phantom utilized a custom insert with two-insert targets filled with 18F-fluorodeoxyglucose (FDG) diluted water. Spherical target had a diameter of 22 mm and C-shape geometry consisted of an active volume of 26mm axial length and 255-degree partial annulus with major and minor radii of 27 mm x 12 mm, respectively. The targets were filled with an activity concentration of 9.3 kBq/ml to achieve a target to background ratio of 8:1 simulating the target uptake with hot background in a patient. The phantom was imaged within 30 minutes of preparation on the X1 PET and then immediately transferred to a PET/CT Biograph scanner for a second image. This experiment was repeated three times in separate days to assess reproducibility of PET acquisition. The PET images were visually assessed by the radiation oncologists to ensure the PET signal is within PTV. The PET signals of the targets were also evaluated using the full-width-half-max (FWHM) calculated on left-right (RL), anterior-posterior (AP) and superior-inferior (SI) directions for both targets and compared to the physical dimensions of the cavities. Results The imaging acquisition tests were successfully completed on all three days for both systems, as per the prescribed workflow. In all experiments, radiation oncologists verified that the PET avid area were within the PTV in both systems. Average FWHM results of the 22 mm spherical target on X1 PET were 24.5, 21.7 and 17.0 mm as compared to 20.0, 20.0 and 18.2 mm for PET/CT in LR, AP and SI directions, respectively. For the C-shape object X1 PET resulted in 18.8, 19.5, 21.8 and 25.0 mm whereas the PET/CT resulted in 12.7, 13.2, 13.2 and 24.7 mm FWHMs for the two RL C-shape legs, AP and SI directions, respectively as compared to 15-, 15-, 15- and 26-mm physical dimensions. Conclusion The X1 PET subsystem - which utilizes dual PET arcs and a sinogram image reconstruction algorithm in order to integrate into a LINAC architecture - generates images that are qualitatively comparable to a traditional PET scanner. In general, X1 FWHM results were slightly larger and PET/CT ones were smaller than the physical dimensions in axial plane. Both systems slightly underestimated the target size in SI direction.

Purpose/Objective(s) Stereotactic ablative radiotherapy (SABR) is an effective treatment for lung tumors, but can result in toxicity such as chest wall pain and life-threatening damage to central lung structures. We hypothesized that while larger tumors require higher dose, small tumors up to 10cc in volume can be controlled with biologically effective dose Materials/Methods Patients in three groups were enrolled: initial diagnosis of non-small cell lung cancer (NSCLC), AJCC 7th edition stage T1-3 N0 M0 (group 1); new primary NSCLC with history of NSCLC, or multiple synchronously diagnosed NSCLCs (group 2); and lung metastases from NSCLC or another primary site (group 3). Up to four tumors could be treated with once-daily SABR. There were six dose/fractionation schedules used, depending on gross tumor volume (≤10cc, 10-30cc, > 30cc) and location (peripheral vs. central). Larger tumors received higher dose and central tumors generally received lower dose per fraction. Dose ranged from 25Gy in one fraction for 0-10cc peripheral tumors to 60Gy in 8 fractions for > 30cc central tumors. Colorectal cancer metastases were treated to higher dose, at least 50Gy in 4 fractions. The primary endpoint was per-group cumulative incidence of local recurrence at 1 year (recurrence of treated tumor within same lobe), with distant recurrence and death as competing risks. Treated tumor recurrence (recurrence with epicenter within 1cm of PTV) and toxicity were also analyzed. Results A total of 217 patients were enrolled from 2011-2018 (some patients were enrolled multiple times). Median age was 72, 59% were male, and 69% were current/former smokers. There were 240 treatment courses and 285 tumors treated (range 1-3 tumors per course). 211 tumors were peripheral and 74 were central. Tumor size distribution was: ≤10cc, 74%; 10-30cc, 19%; > 30cc, 7%. The most common dose was 25Gy in one fraction (158 tumors). Median follow-up was 30 months (range 2-95). Median overall survival was 57 months. Local recurrence data are currently being updated and will be presented at the meeting. The rate of grade 2 or higher pneumonitis was 16/217 (7%) and grade 3 or higher pneumonitis was 3/217 (1%). The rate of grade 2 or higher chest wall pain was 13/217 (6%). One patient had a grade 5 adverse event, developing pulmonary hemorrhage that was possibly related to radiotherapy, 17 months after treatment of a large central NSCLC. Conclusion Individualized SABR to lung cancers resulted in excellent local control and favorable toxicity profile.

Author(s): Gensheimer, Michael F; Yom, Sue S; Soto, Nancy; Dignam, James J; Le, Quynh-Thu; Machtay, Mitchell; Curran, Walter J

Prognostic biomarkers that can reliably predict early disease progression of non-small cell lung cancer (NSCLC) are needed for identifying those patients at high risk for progression, who may benefit from more intensive treatment. In this work, we aimed to identify an imaging signature for predicting progression-free survival (PFS) of locally advanced NSCLC. Methods: This retrospective study included 82 patients with stage III NSCLC treated with definitive chemoradiotherapy for whom both baseline and mid-treatment PET/CT scans were performed. They were randomly placed into two groups: training cohort (n=41) and testing cohort (n=41). All primary tumors and involved lymph nodes were delineated. Forty-five quantitative imaging features were extracted to characterize the tumors and involved nodes at baseline and mid-treatment as well as differences between two scans performed at these two points. An imaging signature was developed to predict PFS by fitting an L1-regularized Cox regression model. Results: The final imaging signature consisted of three imaging features: the baseline tumor volume, the baseline maximum distance between involved nodes, and the change in maximum distance between the primary tumor and involved nodes measured at two time points. According to multivariate analysis, the imaging model was an independent prognostic factor for PFS in both the training (hazard ratio [HR], 1.14, 95% confidence interval [CI], 1.04-1.24; P = 0.003), and testing (HR, 1.21, 95% CI, 1.10-1.33; P = 0.048) cohorts. The imaging signature stratified patients into low- and high-risk groups, with 2-year PFS rates of 61.9% and 33.2%, respectively (P = 0.004 [log-rank test]; HR, 4.13, 95% CI, 1.42-11.70) in the training cohort, as well as 43.8% and 22.6%, respectively (P = 0.006 [log-rank test]; HR, 3.45, 95% CI, 1.35-8.83) in the testing cohort. In both cohorts, the imaging signature significantly outperformed conventional imaging metrics, including tumor volume and SUVmax value (C-indices: 0.77-0.79 for imaging signature, and 0.53-0.73 for conventional metrics). Conclusions: Evaluation of early treatment response by combining primary tumor and nodal imaging characteristics may improve the prediction of PFS of locally advanced NSCLC patients.

Purpose/Objective(s) Contouring organs at risk (OARs) is a key step of radiotherapy treatment planning. In this study, a deep learning-based framework was developed to automatize the whole OAR contouring process contributing to the clinical practice. Materials/Methods This auto-segmentation framework consisted of a pre-processing module, a deep learning-based segmentation module, and a post-processing module. We focused on the auto-segmentation in head and neck, and pelvis sites, and CT images with manual OAR contours from 67 and 54 patient cases were collected from our clinical database for the deep learning, respectively. 16 head and neck OARs and 11 pelvis OARs which were contoured in most patient cases were selected for auto-segmentation. In the pre-processing module, the original image data was extracted from the clinical DICOM Image file, and then processed for the segmentation network learning. A U-Net like network was designed in the auto-segmentation module, which was designed to consider the contour consistency among adjacent image slices and speed up the network convergency. Manual OAR contours were used as the ground truth for the network training, and a weighted Dice loss function was utilized to handle the missing organ contours in some patient cases. The segmentation results would be converted to a DCIOM Structure file in the post-processing module for treatment planning. Results The accuracy of the current auto-segmentation framework in terms of the Dice similarity coefficient (DSC) is shown in the table. Once a clinical DICOM Image data is sent to the framework, a corresponding DICOM Structure file would be automatically generated containing the OAR segmentation results. The overall operation takes about 1-2 min per case without the need for additional manual interference. Conclusion A deep learning-based framework was developed for the auto-segmentation of head and neck, and pelvis OARs with promising contouring accuracy and operation time, which is now ready to be applied in our clinical practice. Moreover, this framework is being developed toward a physician-specific framework to comply with the contour style of each radiation oncologist separately.

Purpose/Objective(s) Propagation of contours from high-quality magnetic resonance (MR) images to treatment planning ultrasound (US) images with severe needle artifacts is a challenging task, which can greatly aid the organ or intraprostatic lesion contouring in high dose rate (HDR) prostate brachytherapy. In this study, a deep learning approach was developed to automatize this registration procedure for HDR brachytherapy practice. Materials/Methods With the approval from the institutional review board, the MR and US images with manual prostate contours were collected from our clinical database and processed to construct the deep learning dataset. A deep learning framework based on the weakly-supervised strategy was proposed for the registration of prostate MR images and US images after the needle insertion, which consisted of two segmentation networks and a deformable registration network. Specifically, two 3D V-Net was trained for the prostate segmentation on the MR and US images separately, to generate the weak supervision labels automatically for the registration network training. Besides the image pair, the corresponding prostate probability maps from the segmentation were further fed to the registration network to improve the registration accuracy, and the deformation matrix was the output. A random augmentation method was designed to alleviate the map overfitting problem during the network training. The overlap between the deformed and fixed prostate contours was analyzed to evaluate the registration accuracy. Results Clinical data from 121, 104 and 63 patient cases were collected and utilized for the MR and US image segmentation networks, and the registration network learning, respectively. The mean DSC, center-of-mass (COM) distance, Hausdorff distance (HD) and averaged symmetric surface distance (ASSD) results for the registration of prostate contours were 0.87 ± 0.05, 1.70 ± 0.89 mm, 7.21 ± 2.07 mm, 1.61 ± 0.64 mm, respectively. By providing the prostate probability map from the segmentation to the registration network, as well as applying the random map augmentation method, the evaluation results of the four metrics were all improved, such as an increase of DSC from 0.83 ± 0.08 to 0.86 ± 0.06 and from 0.86 ± 0.06 to 0.87 ± 0.05, respectively. Conclusion This study first worked on the deep learning-based registration about the treatment planning US image for HDR prostate brachytherapy. A novel segmentation-based registration framework was proposed for this multi-modality registration problem, which not only largely saved the labor work on the data preparation, but also improved the registration accuracy. The evaluation results showed the potential of this approach in HDR brachytherapy to automatize current clinical procedures.

Purpose/Objective(s) Surveillance after definitive radiotherapy for oropharyngeal squamous cell carcinoma (OPSCC) is critical to confirm clinical complete response, as a subset of patients develop locoregional recurrences (LRR). The Hopkins Criteria are a standardized qualitative system for assessing response using post-treatment PET/CT and are validated for overall (OS) and progression-free survival (PFS). We hypothesized that the Hopkins Criteria score is independently prognostic for LRR. Materials/Methods We conducted an institutional retrospective study of adult patients with newly-diagnosed node-positive non-metastatic OPSCC treated with definitive (chemo)radiotherapy. Patients were eligible if they completed a pre- and post-radiotherapy PET/CT within 24 weeks after treatment. The primary outcome was the cumulative incidence of locoregional recurrence/persistence. A single board-certified nuclear medicine physician assigned Hopkins scores, which were considered either positive (scores 4-5) or negative (scores 1-3). Results Between 2004-2018, 259 patients were included in this analysis, while an additional 221 were excluded, most commonly due to lack of post-treatment PET/CT [37%] or N0 disease [22%]. Most patients were men (90%), 47% had never smoked, 86% had p16+/HPV+ tumors, and 39% had T3-4 tumors. Nearly all (98%) received concurrent chemotherapy, while 11% received induction chemotherapy. Median clinical follow-up was 40 months. Following radiotherapy, the first post-treatment PET/CT was positive (score 4-5) at the primary site in 32 patients (12%), cervical lymph nodes in 26 patients (10%), and either the primary site or neck in 50 patients (19%). This PET/CT was performed at a median of 3.3 months [range, 1.6-6.0] after radiotherapy. Thirty-four (13%) LRRs were observed. The sensitivity, specificity, negative predictive value, and overall accuracy of the Hopkins score for LRR were 68%, 88%, 95%, and 85%. The 36-month cumulative incidence of LRR was greater among patients with Hopkins scores 4-5 (45% vs. 5%, multivariable-adjusted HR 14.16, P Conclusion There is a lack of consensus reporting standards for response assessment after definitive radiotherapy for OPSCC. The Hopkins Criteria are an easily-implemented standard that appear to offer high accuracy for recurrence and death. These data support an initial PET/CT at least three months after radiotherapy to limit unnecessary diagnostic procedures.

Purpose/Objective(s) To evaluate Biology-guided Radiotherapy (BgRT) plans generated using PET data acquired on the first installation of the Reflexion™ X1 system. BgRT treatment planning optimizes for fluence indirectly by calculating firing filters that best translate the X1 acquired PET image to the fluences that deliver the prescribed radiation dose. During a treatment session, the firing filters are applied to limited time sampled (LTS) PET images to calculate delivery fluences. Materials/Methods An FDG fillable phantom insert that consists of a 22 mm diameter spherical target (PTV1, ball) and a C-shape object (PTV2 with 26mm axial length, 255° partial annulus with major and minor radii of 27 mm and 12 mm) was imaged using a CT Simulator to create planning contours. The water filled insert surrounding these objects was filled with FDG to simulate a warm background. Both targets were filled with FDG, simulating a target to background ratio of 8:1. Two different BgRT treatment plans were generated for PTV1 simulating homogenous uptake, and PTV2 with an inhomogeneous FDG uptake due to necrosis (a cylinder encompassing the C shape PET avid object). A structure that is 12.5 mm away from these targets was used to simulate OAR. Two PET/CT imaging acquisitions were performed on the RefleXion X1 masking the signal outside of each target plus 10 mm margin. Two different BgRT plans were generated prescribing 50 Gy delivered in 5 fractions to each target, separately. The quality of each treatment plan was evaluated based on PTV coverage, Conformity Index at 50% and 80% isodose lines (CI50, CI80) and OAR sparing. Both BgRT plans then delivered to the independent quality assurance tool for patient specific QA verification where the 3%/3 mm gamma criteria was used. Results BgRT plans for both homogenous and heterogenous FDG uptake targets were created successfully. Prescription dose coverages were 94.2% and 96.3% for PTV1 and PTV2, respectively. The maximum OAR doses were 30.9 and 33.0 Gy and average OAR doses were 14.4 and 20.1 Gy for PTV1 and PTV2 plans, respectively. CI50 were 7.7 and 6.8 where CI80 were 2.5 and 2.6 for PTV1 and PTV2 plans. The patient specific QA passed with 100 % for both plans. Conclusion This is the first report of the BgRT plan creation and QA delivery using a clinical installation of the Reflexion X1 system, opening the potential towards BgRT dose delivery using the real time FDG uptake measurements to biologically guide active radiation beam delivery. The X1 system was able to optimize and deliver the BgRT plans for both homogenous and heterogenous FDG uptake scenarios. BgRT is a new paradigm of dose optimization, where a new set of optimization and evaluation approaches will be needed.

Background and purpose Pre- and mid-radiotherapy FDG-PET metrics have been proposed as biomarkers of recurrence and survival in patients treated for stage III non-small cell lung cancer. We evaluated these metrics in patients treated with definitive radiation therapy (RT). We also evaluated outcomes after progression on mid-radiotherapy PET/CT. Material and methods Seventy-seven patients treated with RT with or without chemotherapy were included in this retrospective study. Primary tumor and involved nodes were delineated. PET metrics included metabolic tumor volume (MTV), total lesion glycolysis (TLG), and SUVmax. For mid-radiotherapy PET, both absolute value of these metrics and percentage decrease were analyzed. The influence of PET metrics on time to death, local recurrence, and regional/distant recurrence was assessed using Cox regression. Results 91% of patients had concurrent chemotherapy. Median follow-up was 14 months. None of the PET metrics were associated with overall survival. Several were positively associated with local recurrence: pre-radiotherapy MTV, and mid-radiotherapy MTV and TLG (p = 0.03–0.05). Ratio of mid- to pre-treatment SUVmax was associated with regional/distant recurrence (p = 0.02). 5/77 mid-radiotherapy scans showed early out-of-field progression. All of these patients died. Conclusions Several PET metrics were associated with risk of recurrence. Progression on mid-radiotherapy PET/CT was a poor prognostic factor.

To evaluate the prognostic value and molecular basis of a CT-derived pleural contact index (PCI) in early stage non-small cell lung cancer (NSCLC). We retrospectively analysed seven NSCLC cohorts. A quantitative PCI was defined on CT as the length of tumour-pleura interface normalised by tumour diameter. We evaluated the prognostic value of PCI in a discovery cohort (n = 117) and tested in an external cohort (n = 88) of stage I NSCLC. Additionally, we identified the molecular correlates and built a gene expression-based surrogate of PCI using another cohort of 89 patients. To further evaluate the prognostic relevance, we used four datasets totalling 775 stage I patients with publically available gene expression data and linked survival information. At a cutoff of 0.8, PCI stratified patients for overall survival in both imaging cohorts (log-rank p = 0.0076, 0.0304). Extracellular matrix (ECM) remodelling was enriched among genes associated with PCI (p = 0.0003). The genomic surrogate of PCI remained an independent predictor of overall survival in the gene expression cohorts (hazard ratio: 1.46, p = 0.0007) adjusting for age, gender, and tumour stage. CT-derived pleural contact index is associated with ECM remodelling and may serve as a noninvasive prognostic marker in early stage NSCLC. • A quantitative pleural contact index (PCI) predicts survival in early stage NSCLC. • PCI is associated with extracellular matrix organisation and collagen catabolic process. • A multi-gene surrogate of PCI is an independent predictor of survival. • PCI can be used to noninvasively identify patients with poor prognosis.

We built a natural language processing (NLP) language model that can be used to extract cancer treatment information using structured and unstructured electronic medical records (EMR). Our work appears to be the first that combines EMR and NLP for treatment identification.

To determine if pre-treatment non-target lung FDG-PET uptake predicts for symptomatic radiation pneumonitis (RP) following lung stereotactic ablative radiotherapy (SABR).We reviewed a 258 patient database from our institution to identify 28 patients who experienced symptomatic (grade ⩾ 2) RP after SABR, and compared them to 57 controls who did not develop symptomatic RP. We compared clinical, dosimetric and functional imaging characteristics between the 2 cohorts including pre-treatment non-target lung FDG-PET uptake.Median follow-up time was 26.9 months. Patients who experienced symptomatic RP had significantly higher non-target lung FDG-PET uptake as measured by mean SUV (p 0.0001) than controls. ROC analysis for symptomatic RP revealed area under the curve (AUC) of 0.74, with sensitivity 82.1% and specificity 57.9% with cutoff mean non-target lung SUV 0.56. Predictive value increased (AUC of 0.82) when mean non-target lung SUV was combined with mean lung dose (MLD). We developed a 0-2 point model using these 2 variables, 1 point each for SUV 0.56 or MLD 5.88 Gy equivalent dose in 2 Gy per fraction (EQD2), predictive for symptomatic RP in our cohort with hazard ratio 10.01 for score 2 versus 0 (p 0.001).Patients with elevated pre-SABR non-target lung FDG-PET uptake are at increased risk of symptomatic RP after lung SABR. Our predictive model suggests patients with mean non-target lung SUV 0.56 and MLD 5.88 Gy EQD2 are at highest risk. Our predictive model should be validated in an external cohort before clinical implementation.

We propose a deep learning model - Probabilistic Prognostic Estimates of Survival in Metastatic Cancer Patients (PPES-Met) for estimating short-term life expectancy (>3 months) of the patients by analyzing free-text clinical notes in the electronic medical record, while maintaining the temporal visit sequence. In a single framework, we integrated semantic data mapping and neural embedding technique to produce a text processing method that extracts relevant information from heterogeneous types of clinical notes in an unsupervised manner, and we designed a recurrent neural network to model the temporal dependency of the patient visits. The model was trained on a large dataset (10,293 patients) and validated on a separated dataset (1818 patients). Our method achieved an area under the ROC curve (AUC) of 0.89. To provide explain-ability, we developed an interactive graphical tool that may improve physician understanding of the basis for the model’s predictions. The high accuracy and explain-ability of the PPES-Met model may enable our model to be used as a decision support tool to personalize metastatic cancer treatment and provide valuable assistance to the physicians.

There is currently great interest in applying neural networks to prediction tasks in medicine. It is important for predictive models to be able to use survival data, where each patient has a known follow-up time and event/censoring indicator. This avoids information loss when training the model and enables generation of predicted survival curves. In this paper, we describe a discrete-time survival model that is designed to be used with neural networks, which we refer to as Nnet-survival. The model is trained with the maximum likelihood method using mini-batch stochastic gradient descent (SGD). The use of SGD enables rapid convergence and application to large datasets that do not fit in memory. The model is flexible, so that the baseline hazard rate and the effect of the input data on hazard probability can vary with follow-up time. It has been implemented in the Keras deep learning framework, and source code for the model and several examples is available online. We demonstrate the performance of the model on both simulated and real data and compare it to existing models Cox-nnet and Deepsurv.

Identifying molecular residual disease (MRD) after treatment of localized lung cancer could facilitate early intervention and personalization of adjuvant therapies. Here, we apply cancer personalized profiling by deep sequencing (CAPP-seq) circulating tumor DNA (ctDNA) analysis to 255 samples from 40 patients treated with curative intent for stage I–III lung cancer and 54 healthy adults. In 94% of evaluable patients experiencing recurrence, ctDNA was detectable in the first posttreatment blood sample, indicating reliable identification of MRD. Posttreatment ctDNA detection preceded radiographic progression in 72% of patients by a median of 5.2 months, and 53% of patients harbored ctDNA mutation profiles associated with favorable responses to tyrosine kinase inhibitors or immune checkpoint blockade. Collectively, these results indicate that ctDNA MRD in patients with lung cancer can be accurately detected using CAPP-seq and may allow personalized adjuvant treatment while disease burden is lowest. Significance: This study shows that ctDNA analysis can robustly identify posttreatment MRD in patients with localized lung cancer, identifying residual/recurrent disease earlier than standard-of-care radiologic imaging, and thus could facilitate personalized adjuvant treatment at early time points when disease burden is lowest. Cancer Discov; 7(12); 1394–403. ©2017 AACR. See related commentary by Comino-Mendez and Turner, p. 1368. This article is highlighted in the In This Issue feature, p. 1355

Purpose To determine if regional ventilation within irradiated lung volume predicts change in pulmonary function test (PFT) measurements after stereotactic ablative radiotherapy (SABR) of lung tumors. Methods We retrospectively identified 27 patients treated from 2007 to 2014 at our institution who received: (1) SABR without prior thoracic radiation; (2) pre-treatment 4-dimensional computed tomography (4-D CT) imaging; (3) pre- and post-SABR PFTs 20BED3 ) as the quotient of mean ventilation (mean Jacobian-based per-voxel volume change on deformably registered inhale/exhale 4-D CT phases) within the 20Gy biologically effective dose ( α / β =3Gy) isodose volume and that of the total lung volume (TLV). Results Most patients had moderate to very severe COPD by GOLD criteria ( n =19, 70.1%). Higher VR 20BED3 significantly predicted worse change in Forced Expiratory Volume/s normalized by baseline value (ΔFEV 1 /FEV 1pre , p =0.04); n =7 had VR 20BED3 >1 (high regional ventilation) and worse ΔFEV 1 /FEV 1pre (median=−0.16, range=−0.230 to −0.20). Five had VR 20BED3 1 /FEV 1pre (median=0.13, range=0.07 to 0.20). In a multivariable linear model, increasing VR 20BED3 and time to post-SABR PFT predicted decreasing ΔFEV 1 /FEV 1pre ( R 2 =0.25, p =0.03). Conclusions After SABR to high versus low functioning lung regions, we found worsened or improved global pulmonary function, respectively. If pre-SABR VR 20BED3 is validated as a predictor of eventual post-SABR PFT in larger studies, it may be used for individualized treatment planning to preserve or even improve pulmonary function after SABR.

Purpose To develop an intratumor partitioning framework for identifying high-risk subregions from 18 F-fluorodeoxyglucose positron emission tomography (FDG-PET) and computed tomography (CT) imaging and to test whether tumor burden associated with the high-risk subregions is prognostic of outcomes in lung cancer. Methods and Materials In this institutional review board–approved retrospective study, we analyzed the pretreatment FDG-PET and CT scans of 44 lung cancer patients treated with radiation therapy. A novel, intratumor partitioning method was developed, based on a 2-stage clustering process: first at the patient level, each tumor was over-segmented into many superpixels by k-means clustering of integrated PET and CT images; next, tumor subregions were identified by merging previously defined superpixels via population-level hierarchical clustering. The volume associated with each of the subregions was evaluated using Kaplan-Meier analysis regarding its prognostic capability in predicting overall survival (OS) and out-of-field progression (OFP). Results Three spatially distinct subregions were identified within each tumor that were highly robust to uncertainty in PET/CT co-registration. Among these, the volume of the most metabolically active and metabolically heterogeneous solid component of the tumor was predictive of OS and OFP on the entire cohort, with a concordance index or CI of 0.66-0.67. When restricting the analysis to patients with stage III disease (n=32), the same subregion achieved an even higher CI of 0.75 (hazard ratio 3.93, log-rank P =.002) for predicting OS, and a CI of 0.76 (hazard ratio 4.84, log-rank P =.002) for predicting OFP. In comparison, conventional imaging markers, including tumor volume, maximum standardized uptake value, and metabolic tumor volume using threshold of 50% standardized uptake value maximum, were not predictive of OS or OFP, with CI mostly below 0.60 (log-rank P >.05). Conclusion We propose a robust intratumor partitioning method to identify clinically relevant, high-risk subregions in lung cancer. We envision that this approach will be applicable to identifying useful imaging biomarkers in many cancer types.

BACKGROUND: Oncologists use patients’ life expectancy to guide decisions and may benefit from a tool that accurately predicts prognosis. Existing prognostic models generally use only a few predictor variables. We used an electronic medical record dataset to train a prognostic model for patients with metastatic cancer. METHODS: The model was trained and tested using 12 588 patients treated for metastatic cancer in the Stanford Health Care system from 2008 to 2017. Data sources included provider note text, labs, vital signs, procedures, medication orders, and diagnosis codes. Patients were divided randomly into a training set used to fit the model coefficients and a test set used to evaluate model performance (80%/20% split). A regularized Cox model with 4126 predictor variables was used. A landmarking approach was used due to the multiple observations per patient, with t(0) set to the time of metastatic cancer diagnosis. Performance was also evaluated using 399 palliative radiation courses in test set patients. RESULTS: The C-index for overall survival was 0.786 in the test set (averaged across landmark times). For palliative radiation courses, the C-index was 0.745 (95% confidence interval [CI] = 0.715 to 0.775) compared with 0.635 (95% CI = 0.601 to 0.669) for a published model using performance status, primary tumor site, and treated site (two-sided P

Background: Complete genome sequences provide many new characters suitable for studying phylogenetic relationships. The limitations of the single sequence-based phylogenetic reconstruction prompted the efforts to build trees based on genome-wide properties, such as the fraction of shared orthologous genes or conservation of adjoining gene pairs. Gene content-based phylogenies, however, have their own biases: most notably, differential losses and horizontal transfers of genes interfere with phylogenetic signal, each in their own way, and special measures need to be taken to eliminate these types of noise. Results: We expand the repertoire of genome-wide traits available for phylogeny building, by developing a practical approach for measuring local gene conservation in two genomes. We counted the number of orthologous genes shared by chromosomal neighborhoods (“bins”), and built the phylogeny of 63 prokaryotic genomes on this basis. The tree correctly resolved all well-established clades, and also suggested the monophyly of firmicutes, which tend to be split in other genome-based trees. Conclusions: Our measure of local gene order conservation extracts strong phylogenetic signal. This new measure appears to be substantially resistant to the observed instances of gene loss and horizontal transfer, two evolutionary forces which can cause systematic biases in the genome-based phylogenies.