A Competition, Benchmark, Code, and Data for Using Artificial Intelligence to Detect Lesions in Digital Breast Tomosynthesis.

Importance: An accurate and robust artificial intelligence (AI) algorithm for detecting cancer in digital breast tomosynthesis (DBT) could significantly improve detection accuracy and reduce health care costs worldwide. Objectives: To make training and evaluation data for the development of AI algorithms for DBT analysis available, to develop well-defined benchmarks, and to create publicly available code for existing methods. Design, Setting, and Participants: This diagnostic study is based on a multi-institutional international grand challenge in which research teams developed algorithms to detect lesions in DBT. A data set of 22 032 reconstructed DBT volumes was made available to research teams. Phase 1, in which teams were provided 700 scans from the training set, 120 from the validation set, and 180 from the test set, took place from December 2020 to January 2021, and phase 2, in which teams were given the full data set, took place from May to July 2021. Main Outcomes and Measures: The overall performance was evaluated by mean sensitivity for biopsied lesions using only DBT volumes with biopsied lesions; ties were broken by including all DBT volumes. Results: A total of 8 teams participated in the challenge. The team with the highest mean sensitivity for biopsied lesions was the NYU B-Team, with 0.957 (95% CI, 0.924-0.984), and the second-place team, ZeDuS, had a mean sensitivity of 0.926 (95% CI, 0.881-0.964). When the results were aggregated, the mean sensitivity for all submitted algorithms was 0.879; for only those who participated in phase 2, it was 0.926. Conclusions and Relevance: In this diagnostic study, an international competition produced algorithms with high sensitivity for using AI to detect lesions on DBT images. A standardized performance benchmark for the detection task using publicly available clinical imaging data was released, with detailed descriptions and analyses of submitted algorithms accompanied by a public release of their predictions and code for selected methods. These resources will serve as a foundation for future research on computer-assisted diagnosis methods for DBT, significantly lowering the barrier of entry for new researchers.

Automatic Shuttlecock Fall Detection System in or out of a Court in Badminton Games-Challenges, Problems, and Solutions from a Practical Point of View.

We built an Instant Review System (IRS) for badminton, also named a Challenge System. It allows players to verify linesmen in/out decisions and makes the game fairer. Elements such as lighting, the influence of air-conditioning on the flight trajectory, or the moving mats can significantly impact the final in/out decision. Due to the construction of the shuttlecock, it behaves differently during the flight than, for example, a tennis ball. This publication discusses the problems we encountered during our work with the proposed solution. We present the evolution of the system's architecture: the first version with the cameras mounted above the court and placed around the court close to the lines, tracking the shuttlecock in 3D; and the second, improved version with cameras placed only around the court, without 3D reconstruction. We used our system during the BWF World Senior Badminton Championships in Katowice. We present the system's results from this tournament and compare them with linesmen's decisions. We describe the system's verification process by the Badminton World Federation and Polish Badminton Federation and discuss evaluation methods for such systems. Our solution is comparable to the commercial product used in the biggest badminton tournaments in regard to processing time and accuracy. Still, our architecture and algorithms make installing it much easier and faster, making the system more adaptive, reliable, flexible, and universal in relation to the practical requirements of sports halls.

Checklist for responsible deep learning modeling of medical images based on COVID-19 detection studies.

The sudden outbreak and uncontrolled spread of COVID-19 disease is one of the most important global problems today. In a short period of time, it has led to the development of many deep neural network models for COVID-19 detection with modules for explainability. In this work, we carry out a systematic analysis of various aspects of proposed models. Our analysis revealed numerous mistakes made at different stages of data acquisition, model development, and explanation construction. In this work, we overview the approaches proposed in the surveyed Machine Learning articles and indicate typical errors emerging from the lack of deep understanding of the radiography domain. We present the perspective of both: experts in the field - radiologists and deep learning engineers dealing with model explanations. The final result is a proposed checklist with the minimum conditions to be met by a reliable COVID-19 diagnostic model.

Effect of domain knowledge encoding in CNN model architecture-a prostate cancer study using mpMRI images.

BACKGROUND: Prostate cancer is one of the most common cancers worldwide. Currently, convolution neural networks (CNNs) are achieving remarkable success in various computer vision tasks, and in medical imaging research. Various CNN architectures and methodologies have been applied in the field of prostate cancer diagnosis. In this work, we evaluate the impact of the adaptation of a state-of-the-art CNN architecture on domain knowledge related to problems in the diagnosis of prostate cancer. The architecture of the final CNN model was optimised on the basis of the Prostate Imaging Reporting and Data System (PI-RADS) standard, which is currently the best available indicator in the acquisition, interpretation, and reporting of prostate multi-parametric magnetic resonance imaging (mpMRI) examinations. METHODS: A dataset containing 330 suspicious findings identified using mpMRI was used. Two CNN models were subjected to comparative analysis. Both implement the concept of decision-level fusion for mpMRI data, providing a separate network for each multi-parametric series. The first model implements a simple fusion of multi-parametric features to formulate the final decision. The architecture of the second model reflects the diagnostic pathway of PI-RADS methodology, using information about a lesion's primary anatomic location within the prostate gland. Both networks were experimentally tuned to successfully classify prostate cancer changes. RESULTS: The optimised knowledge-encoded model achieved slightly better classification results compared with the traditional model architecture (AUC = 0.84 vs. AUC = 0.82). We found the proposed model to achieve convergence significantly faster. CONCLUSIONS: The final knowledge-encoded CNN model provided more stable learning performance and faster convergence to optimal diagnostic accuracy. The results fail to demonstrate that PI-RADS-based modelling of CNN architecture can significantly improve performance of prostate cancer recognition using mpMRI.

Neuroimaging of Cerebral Venous Thrombosis (CVT) - Old Dilemma and the New Diagnostic Methods.

BACKGROUND: Cerebral venous thrombosis is a relatively uncommon neurologic disorder that is potentially reversible with prompt diagnosis and appropriate medical care. The pathogenesis is multifactorial and the disease may occur at any age. CVT is often associated with nonspecific symptoms. Radiologists play a crucial role in patient care by providing early diagnosis through interpretation of imaging studies. Underdiagnosis or misdiagnosis can increase the risk of severe complications, including hemorrhagic stroke or death. The purpose of this study is to investigate radiological and clinical characteristics of cerebral venous thrombosis (CVT) based on material from 34 patients under care of our hospital. MATERIAL/METHODS: A total of 34 patients were diagnosed with CVT from August 2009 until March 2015. A clinical and radiological database of patients with final diagnosis of CVT was analyzed. RESULTS: Patient group included 22 women and 12 men at a mean age of 48.7 years (ranging from 27 to 77 years). In the study group 8 patients (23.5%) suffered from hemorrhagic infarction, whereas 16 patients (47%) were diagnosed with venous infarction without hemorrhage. Thirty patients (88%) had transverse sinus thrombosis. CONCLUSIONS: According to our study, CVT was more prevalent in women. Transverse sinus was the most common location. Among all age groups, the highest prevalence was seen in the fifth decade (n=14). Contrast-enhanced CT and MR venography were the most sensitive imaging modalities.

Recovery of CT stroke hypodensity--An adaptive variational approach.

The present research was directed to effective image restoration with the extraction of ischemic edema signs. Computerized support of hyperacute stroke diagnosis based on routinely used computerized tomography (CT) scans was optimized to visualize the infarct extent more precisely. In particular, a beneficial support of time-limited appropriate decision of whether to treat the patient by thrombolysis is expected. Because of a limited accuracy in determining the area of core infarction, particularly in the early hours of symptoms' onset, a variational approach to sensed data recovery was applied. Proposed methodology adjusts fidelity norms and regularization priors integrated with simulated sensing procedures in a compressed sensing framework. Experimental study confirmed almost perfect recognition of ischemic stroke in a test set of over 500 CT scans.

Hypodensity extractor: a phantom study.

We report on the extraction procedures of low-contrast symptomatic hypodensity optimized for a computed tomography-based diagnosis. The specific application is brain imaging with enhanced perception of hypodense areas which are direct symptoms of acute ischemia. A standard low-contrast phantom, as commonly employed in dosimetry and imaging quality evaluation, was used to derive numeric criteria for assessing the extraction effectiveness. Our proposed procedure is based on multiscale analysis of the image data expanded over the frames of wavelets, curvelets or complex wavelets, followed by nonlinear approximation of the symptom signatures. Apparent subtle density changes in the phantom were evaluated using computational metrics and subjective ratings. We discuss the advantages and disadvantages of our proposed optimized hypodensity extraction procedures.

Multiscale extraction of hypodensity in hyperacute stroke.

BACKGROUND: Stroke is a syndrome characterized by a sudden neurological deficit caused by cerebral ischemia. Computed tomography (CT) plays a crucial role in the evaluation of stroke patients even though it is not sufficient enough to extract the hypodense area corresponding to the infracted cerebral tissues in hyperacute stage. Careful selection of patients for thrombolytic therapy is fundamental to improve safety and efficacy; therefore the authors propose an additional, wavelet-based post-processing method for extracting hypodensity in CT scans in hyperacute stroke patients. MATERIAL/METHODS: A retrospective evaluation of 52 sets of examinations conducted in patients admitted with symptoms suggestive of stroke was undertaken by four radiologists unaware of the final clinical findings. All of the selected cases were considered as having no direct signs of hyperacute ischemia in the localization corresponding with clinical manifestation and follow-up studies. In the first stage, only CTs performed at admission were evaluated; a month later the same scans were reevaluated with additional use of a "stroke monitor". All sets were evaluated independently. Follow-up CT exam and/or the clinical picture confirmed or excluded the diagnosis. RESULTS: Higher AUC values were found for the "stroke monitor"-aided radiological diagnosis for all the radiologists and the differences were significant for all subgroups (p<0.05) apart from the subgroup in which CT scans of patients with significant cerebral atrophy were excluded. CONCLUSIONS: Combining the results of CT and the "stroke monitor" provided a better diagnosis of stroke, especially in atrophic brains. Planned prospective studies will allow evaluation of the impact on the further treatment of hyperacute stroke patients.

Vector quality measure of lossy compressed medical images.

A numerical measure, which is able to predict diagnostic accuracy rather than subjective quality, is required for compressed medical image assessment. The objective of this study is to present a proposal for a new vector measure of image quality, reflecting diagnostic accuracy. Construction of such measure includes the formation of a diagnostic quality pattern based on the subjective ratings of local image features playing an essential role in the detection and classification of any lesion. Experimental results contain the opinions of 9 radiologists: 2 test designers and 7 observers who rated digital mammograms. The correlation coefficient between the numerical equivalent of the vector measure and subjective pattern is over 0.9.