January 2021
Eigenrank by Committee: Von-Neumann Entropy Based Data Subset Selection and Failure Prediction for Deep Learning Based Medical Image Segmentation.Manual delineation of anatomy on existing images is the basis of developing deep learning algorithms for medical image segmentation. However, manual segmentation is tedious. It is also expensive because clinician effort is necessary to ensure correctness of delineation. Consequently most algorithm development is based on a tiny fraction of the vast amount of imaging data collected at a medical center. Thus, selection of a subset of images from hospital databases for manual delineation - so that algorithms trained on such data are accurate and tolerant to variation, becomes an important challenge. We address this challenge using a novel algorithm. The proposed algorithm named 'Eigenrank by Committee' (EBC) first computes the degree of disagreement between segmentations generated by each DL model in a committee. Then, it iteratively adds to the committee, a DL model trained on cases where the disagreement is maximal. The disagreement between segmentations is quantified by the maximum eigenvalue of a Dice coefficient disagreement matrix a measure closely related to the Von Neumann entropy. We use EBC for selecting data subsets for manual labeling from a larger database of spinal canal segmentations as well as intervertebral disk segmentations. U-Nets trained on these subsets are used to generate segmentations on the remaining data. Similar sized data subsets are also randomly sampled from the respective databases, and U-Nets are trained on these random subsets as well. We found that U-Nets trained using data subsets selected by EBC, generate segmentations with higher average Dice coefficients on the rest of the database than U-Nets trained using random sampling (p < 0.05 using t-tests comparing averages). Furthermore, U-Nets trained using data subsets selected by EBC generate segmentations with a distribution of Dice coefficients that demonstrate significantly (p < 0.05 using Bartlett's test) lower variance in comparison to U-Nets trained using random sampling for all datasets. We believe that this lower variance indicates that U-Nets trained with EBC are more robust than U-Nets trained with random sampling.
January 2021
[Future of Cerebral Aneurysm Treatment].There has been an increasing role in the low invasive endovascular treatment of intracranial aneurysms. In addition to the detachable coils, the development of intracranial stents that are capable of repairing the parent artery itself has induced a significant treatment paradigm shift from open surgical to endovascular intervention. Recent evidence suggests that chronic inflammation plays a critical role in the process of intracranial aneurysm formation and rupture. It is, therefore, a natural evolution to seek drug treatments for intracranial aneurysms for growth or rupture prevention rather than any mechanical intervention. The authors review the current preclinical efforts on aneurysm drug treatments and prospective. Also covered is an emerging technology such as robotic endovascular treatment. The robotic system is capable of performing a subset of endovascular procedures such as stent-assisted aneurysm coiling. Although a lot of work needs to be done, remote health care is no longer science fiction.
January 2021
Call for a New Radiology Subspecialty in Imaging-Based Screening.Imaging-based screening has become a critical component of preventive care medicine, growing immensely over the past 50 years. Radiologists are at the center of this public health practice—we are the imaging experts—and yet we are underrepresented in the decision-making process that directs national screening practices. These decisions are largely made by primary care professionals and epidemiologists, who lack expertise in imaging. Here are two possible reasons for this: (1) Radiologists currently have minimal training in epidemiology and evidence development related to imaging-based screening, and (2) radiologists may be viewed as fundamentally biased in favor of imaging, resulting in a daily incentive toward more screening. As a solution, we propose a new radiology subspecialty—screening radiology—to help close the educational gap, untangle advocacy from science, and enable more effective radiology leadership in screening.
January 2021
Cardiac Magnetic Resonance Quantification of Structure-Function Relationships in Heart Failure.Classification of heart failure is based on the left ventricular ejection fraction (EF): preserved EF, midrange EF, and reduced EF. There remains an unmet need for further heart failure phenotyping of ventricular structure-function relationships. Because of high spatiotemporal resolution, cardiac magnetic resonance (CMR) remains the reference modality for quantification of ventricular contractile function. The authors aim to highlight novel frameworks, including theranostic use of ferumoxytol, to enable more efficient evaluation of ventricular function in heart failure patients who are also frequently anemic, and to discuss emerging quantitative CMR approaches for evaluation of ventricular structure-function relationships in heart failure.
January 2021
PI-RADS Version 2.1: A Critical Review, From the AJR Special Series on Radiology Reporting and Data Systems.PI-RADS version 2.1 updates the technical parameters for multiparametric MRI (mpMRI) of the prostate and revises the imaging interpretation criteria while maintaining the framework introduced in version 2. These changes have been considered an improvement, although some issues remain unresolved, and new issues have emerged. Areas for improvement discussed in this review include the need for more detailed mpMRI protocols with optimization for 1.5-T and 3-T systems; lack of validation of revised transition zone interpretation criteria and need for clarifications of the revised DWI and dynamic contrast-enhanced imaging criteria and central zone (CZ) assessment; the need for systematic evaluation and reporting of background changes in signal intensity in the prostate that can negatively affect cancer detection; creation of a new category for lesions that do not fit into the PI-RADS assessment categories (i.e., PI-RADS M category); inclusion of quantitative parameters beyond size to evaluate lesion aggressiveness; adjustments to the structured report template, including standardized assessment of the risk of extraprostatic extension; development of parameters for image quality and performance control; and suggestions for expansion of the system to other indications (e.g., active surveillance and recurrence).
