Search

Background: To address the numerous unmeet clinical needs, in recent years several Machine Learning models applied to medical images and clinical data have been introduced and developed. Even when they achieve encouraging results, they lack evolutionary progression, thus perpetuating their status as autonomous entities. We postulated that different algorithms which have been proposed in the literature to address the same diagnostic task, can be aggregated to enhance classification performance. We suggested a proof of concept to define an ensemble approach useful for integrating different algorithms proposed to solve the same clinical task. Methods: The proposed approach was developed starting from a public database consisting of radiomic features extracted from CT images relating to 535 patients suffering from lung cancer. Seven algorithms were trained independently by participants in the AI4MP working group on Artificial Intelligence of the Italian Association of Physics in Medicine to discriminate metastatic from non-metastatic patients. The classification scores generated by these algorithms are used to train SVM classifier. The Explainable Artificial Intelligence approach is applied to the final model. The ensemble model was validated following an 80–20 hold-out and leave-one-out scheme on the training set. Results: Compared to individual algorithms, a more accurate result was achieved. On the independent test the ensemble model achieved an accuracy of 0.78, a F1-score of 0.57 and a log-loss of 0.49. Shapley values representing the contribution of each algorithm to the final classification result of the ensemble model were calculated. This information represents an added value for the end user useful for evaluating the appropriateness of the classification result on a particular case. It also allows us to evaluate on a global level which methodological approaches of the individual algorithms are likely to have the most impact. Conclusion: Our proposal represents an innovative approach useful for integrating different algorithms that populate the literature and which lays the foundations for future evaluations in broader application scenarios. [ABSTRACT FROM AUTHOR]

This topical review focuses on Patient-Specific Quality Assurance (PSQA) approaches to stereotactic body radiation therapy (SBRT). SBRT requires stricter accuracy than standard radiation therapy due to the high dose per fraction and the limited number of fractions. The review considered various PSQA methods reported in 36 articles between 01/2010 and 07/2022 for SBRT treatment. In particular comparison among devices and devices designed for SBRT, sensitivity and resolution, verification methodology, gamma analysis were specifically considered. The review identified a list of essential data needed to reproduce the results in other clinics, highlighted the partial miss of data reported in scientific papers, and formulated recommendations for successful implementation of a PSQA protocol. [ABSTRACT FROM AUTHOR]

[Display omitted] • AI4MP-Challenge is the first AIFM multicentric experience on machine learning. • The main objective is to improve knowledge and skills of medical physicists on machine learning. • Encountered pitfalls: violation of independence assumption, computation errors, data imbalance. • Providing both cross-validation and an independent test helps to detect implementation issue. • The exclusion of non-robust features does not allow to significantly increase model stability. A novel and unconventional approach to a machine learning challenge was designed to spread knowledge, identify robust methods and highlight potential pitfalls about machine learning within the Medical Physics community. A public dataset comprising 41 radiomic features and 535 patients was employed to assess the potential of radiomics in distinguishing between primary lung tumors and metastases. Each participant developed two classification models using: (i) all features (base model); (ii) only robust features (robust model). Both models were validated with cross-validation and on unseen data. The population stability index (PSI) was used as diagnostic metric for implementation issues. Performance was compared to reference. Base and robust models were compared in terms of performance and stability (coefficient of variation (CoV) of prediction probabilities). PSI detected potential implementation errors in 70 % of models. The dataset exhibited strong imbalance. The average Gmean (i.e. an appropriate metric for imbalance) among all participants was 0.67 ± 0.01, significantly higher than reference Gmean of 0.50 ± 0.04. Robust models performances were slightly worse than base models (p < 0.05). Regarding stability, robust models exhibited lower median CoV on training set only. AI4MP-Challenge models overperformed the reference, significantly improving the Gmean. Exclusion of less-robust features did not improve model robustness and it should be avoided when confounding effects are absent. Other methods, like harmonization or data augmentation, should be evaluated. This study demonstrated how the collaborative effort to foster knowledge on machine learning among medical physicists, through interactive sessions and exchange of information among participants, can result in improved models. [ABSTRACT FROM AUTHOR]

Purpose: Diffusion tensor imaging (DTI) is highly sensitive to noise and improvement of radiofrequency coil technology represents a straightforward way for augmenting signal-to-noise ratio (SNR) performance in magnetic resonance imaging (MRI) scanners. The aim of this study was to characterize the dependence of DTI measurements of fractional anisotropy (FA) and mean diffusivity (MD) on the choice of head coil, comparing two head coils with different functional designs and sensitivities. Methods: Fourteen healthy subjects underwent DTI acquisitions at 1.5 T. Every subject was scanned twice, using a standard quadrature birdcage head coil (coil-A) and an eight-channel array head coil (coil-B). FA and MD maps, estimated using both the linear least squares (LLS) and nonlinear least squares (NLLS) methods, were nonlinearly normalized into a standard space. Then, volumetric regions of interest encompassing typical white and gray matter structures [splenium of the corpus callosum (SCC), internal capsule (IC), cerebral peduncles (CP), middle cerebellar peduncles (MCP), globus pallidus (GP), thalamus (TH), caudate (CA), and putamen (PU)] were analyzed. Significant differences and trends of variation in DTI measurements were assessed by the Wilcoxon test for paired samples with and without Bonferroni correction for multiple comparisons, respectively. Results: The overall SNR of coil-B was ∼30% higher than that of coil-A. When comparing DTI measurements (coil-B versus coil-A), mean FA values (SCC, IC, and TH), mean MD values (IC, CP, GP, and TH), FA standard deviation (CP, MCP, GP, and CA), and MD standard deviation (IC, CP, TH, and PU) resulted decreased (significant difference, pcor < 0.05, or trend of variation, puncor < 0.05) in several gray and white matter regions of the human brain. With the exception of CP, the results in terms of revealed significant difference or trend of variation were independent of the method (LLS and NLLS) used for estimating the diffusion tensor. Conclusions: In various gray and white matter structures, the eight-channel array head coil yielded more precise and accurate measurements of DTI derived indices compared to the standard quadrature birdcage head coil. [ABSTRACT FROM AUTHOR]

Abstract: Blood oxygenation level dependent (BOLD) response related to interictal activity was evaluated in a patient with post-traumatic focal epilepsy at repeated continuous electroencephalogram (EEG)-functional magnetic resonance imaging examinations. Lateralized interictal EEG activity induced a main cluster of activation co-localized with the anatomical lesion. Spreading of EEG interictal activity to both frontal lobes evoked bilateral clusters of activation indicating that topography of BOLD response might depend on the spatial distribution of epileptiform activity. [Copyright &y& Elsevier]

Amyotrophic lateral sclerosis is characterized by degeneration of upper and lower motor neurons. Diffusion tensor imaging (DTI) indexes obtained along the corticospinal tracts distinguish ALS patients and control subjects. Diffusion tensor imaging can be estimated from at least 6 diffusion-weighted images; however an acquisition scheme with a higher number of diffusion directions allows a more robust estimation of DTI indexes. The aim of the study was to establish if a higher number of diffusion encoding gradients increases the diagnostic accuracy of DTI in ALS. We studied 18 patients and 16 control subjects acquiring 2 DTI data sets with 6 and 31 gradient orientations. The mean diffusivity and fractional anisotropy values were measured along the corticospinal tract. Mean diffusivity in ALS was significantly increased (P = 0.026) with respect to control subjects in acquisition scheme with 31 but not (P = 0.214) with 6 diffusion-weighting directions. Fractional anisotropy was significantly lower in patients both with 6 (P = 0.0036) and with 31 (P = 0.0004) diffusion-weighting directions (0.538 vs 0.588 and 0.530 vs 0.594). Fractional anisotropy receiver operating characteristic curve analysis showed a higher diagnostic accuracy by using 31 diffusion-weighting direction (85.76%) with respect to 6 directions (79.86%). Diffusion tensor imaging confirms its potentials in diagnosing ALS with a good accuracy; the acquisition scheme with a higher diffusion-weighting directions seems to better discriminate between ALS patients and control subjects.

The rotational variance dependence of diffusion tensor imaging (DTI) derived parameters on the number of diffusion weighting directions (N) has been investigated by several Monte Carlo simulation studies. However, the dependence of fractional anisotropy (FA) and mean diffusivity (MD) maps on N, in terms of accuracy and contrast between different anatomical structures, has not been assessed in detail. This experimental study further investigated in vivo the effect of the number of diffusion weighting directions on DTI maps of FA and MD. Human brain FA and MD maps of six healthy subjects were acquired at 1.5T with varying N (6, 11, 19, 27, 55). Then, FA and MD mean values in high (FAH, MDH) and low (FAL, MDL) anisotropy segmented brain regions were measured. Moreover, the contrast-to-signal variance ratio (CVRFA, CVRMD) between the main white matter and the surrounding regions was calculated. Analysis of variance showed that FAL, FAH and CVRFA significantly (p < 0.05) depend on N. In particular, FAL decreased (6%-11%) with N, whereas FAH (1.6%-2.5%) and CVRFA (4%-6.5%) increased with N. MDL, MDH and CVRMD did not significantly (p>0.05) depend on N. Unlike MD values, FA values significantly vary with N. It is noteworthy that the observed variation is opposite in low and high anisotropic regions. In clinical studies, the effect of N may represent a confounding variable for anisotropy measurements and the employment of DTI acquisition schemes with high N (> 20) allows an increased CVR and a better visualization of white matter structures in FA maps.