Deep learning for computer-aided abnormalities classification in digital mammogram: A data-centric perspective.

Breast cancer is the most common type of cancer in women, and early abnormality detection using mammography can significantly improve breast cancer survival rates. Diverse datasets are required to improve the training and validation of deep learning (DL) systems for autonomous breast cancer diagnosis. However, only a small number of mammography datasets are publicly available. This constraint has created challenges when comparing different DL models using the same dataset. The primary contribution of this study is the comprehensive description of a selection of currently available public mammography datasets. The information available on publicly accessible datasets is summarized and their usability reviewed to enable more effective models to be developed for breast cancer detection and to improve understanding of existing models trained using these datasets. This study aims to bridge the existing knowledge gap by offering researchers and practitioners a valuable resource to develop and assess DL models in breast cancer diagnosis.

Impact of multi-source data augmentation on performance of convolutional neural networks for abnormality classification in mammography.

Introduction: To date, most mammography-related AI models have been trained using either film or digital mammogram datasets with little overlap. We investigated whether or not combining film and digital mammography during training will help or hinder modern models designed for use on digital mammograms. Methods: To this end, a total of six binary classifiers were trained for comparison. The first three classifiers were trained using images only from Emory Breast Imaging Dataset (EMBED) using ResNet50, ResNet101, and ResNet152 architectures. The next three classifiers were trained using images from EMBED, Curated Breast Imaging Subset of Digital Database for Screening Mammography (CBIS-DDSM), and Digital Database for Screening Mammography (DDSM) datasets. All six models were tested only on digital mammograms from EMBED. Results: The results showed that performance degradation to the customized ResNet models was statistically significant overall when EMBED dataset was augmented with CBIS-DDSM/DDSM. While the performance degradation was observed in all racial subgroups, some races are subject to more severe performance drop as compared to other races. Discussion: The degradation may potentially be due to ( 1) a mismatch in features between film-based and digital mammograms ( 2) a mismatch in pathologic and radiological information. In conclusion, use of both film and digital mammography during training may hinder modern models designed for breast cancer screening. Caution is required when combining film-based and digital mammograms or when utilizing pathologic and radiological information simultaneously.

The EMory BrEast imaging Dataset (EMBED): A Racially Diverse, Granular Dataset of 3.4 Million Screening and Diagnostic Mammographic Images.

Supplemental material is available for this article. Keywords: Mammography, Breast, Machine Learning © RSNA, 2023.

Complete and Resilient Documentation for Operational Medical Environments Leveraging Mobile Hands-free Technology in a Systems Approach: Experimental Study.

BACKGROUND: Prehospitalization documentation is a challenging task and prone to loss of information, as paramedics operate under disruptive environments requiring their constant attention to the patients. OBJECTIVE: The aim of this study is to develop a mobile platform for hands-free prehospitalization documentation to assist first responders in operational medical environments by aggregating all existing solutions for noise resiliency and domain adaptation. METHODS: The platform was built to extract meaningful medical information from the real-time audio streaming at the point of injury and transmit complete documentation to a field hospital prior to patient arrival. To this end, the state-of-the-art automatic speech recognition (ASR) solutions with the following modular improvements were thoroughly explored: noise-resilient ASR, multi-style training, customized lexicon, and speech enhancement. The development of the platform was strictly guided by qualitative research and simulation-based evaluation to address the relevant challenges through progressive improvements at every process step of the end-to-end solution. The primary performance metrics included medical word error rate (WER) in machine-transcribed text output and an F1 score calculated by comparing the autogenerated documentation to manual documentation by physicians. RESULTS: The total number of 15,139 individual words necessary for completing the documentation were identified from all conversations that occurred during the physician-supervised simulation drills. The baseline model presented a suboptimal performance with a WER of 69.85% and an F1 score of 0.611. The noise-resilient ASR, multi-style training, and customized lexicon improved the overall performance; the finalized platform achieved a medical WER of 33.3% and an F1 score of 0.81 when compared to manual documentation. The speech enhancement degraded performance with medical WER increased from 33.3% to 46.33% and the corresponding F1 score decreased from 0.81 to 0.78. All changes in performance were statistically significant (P<.001). CONCLUSIONS: This study presented a fully functional mobile platform for hands-free prehospitalization documentation in operational medical environments and lessons learned from its implementation.

Deep learning for semi-automated unidirectional measurement of lung tumor size in CT.

BACKGROUND: Performing Response Evaluation Criteria in Solid Tumor (RECISTS) measurement is a non-trivial task requiring much expertise and time. A deep learning-based algorithm has the potential to assist with rapid and consistent lesion measurement. PURPOSE: The aim of this study is to develop and evaluate deep learning (DL) algorithm for semi-automated unidirectional CT measurement of lung lesions. METHODS: This retrospective study included 1617 lung CT images from 8 publicly open datasets. A convolutional neural network was trained using 1373 training and validation images annotated by two radiologists. Performance of the DL algorithm was evaluated 244 test images annotated by one radiologist. DL algorithm's measurement consistency with human radiologist was evaluated using Intraclass Correlation Coefficient (ICC) and Bland-Altman plotting. Bonferroni's method was used to analyze difference in their diagnostic behavior, attributed by tumor characteristics. Statistical significance was set at p < 0.05. RESULTS: The DL algorithm yielded ICC score of 0.959 with human radiologist. Bland-Altman plotting suggested 240 (98.4 %) measurements realized within the upper and lower limits of agreement (LOA). Some measurements outside the LOA revealed difference in clinical reasoning between DL algorithm and human radiologist. Overall, the algorithm marginally overestimated the size of lesion by 2.97 % compared to human radiologists. Further investigation indicated tumor characteristics may be associated with the DL algorithm's diagnostic behavior of over or underestimating the lesion size compared to human radiologist. CONCLUSIONS: The DL algorithm for unidirectional measurement of lung tumor size demonstrated excellent agreement with human radiologist.

Intervention to Reduce Interobserver Variability in Computed Tomographic Measurement of Cancer Lesions Among Experienced Radiologists.

While a growing number of research studies have reported the inter-observer variability in computed tomographic (CT) measurements, there are very few interventional studies performed. We aimed to assess whether a peer benchmarking intervention tool may have an influence on reducing interobserver variability in CT measurements and identify possible barriers to the intervention. In this retrospective study, 13 board-certified radiologists repeatedly reviewed 10 CT image sets of lung lesions and hepatic metastases during 3 noncontiguous time periods (T1, T2, T3). Each preselected case contained normal anatomy cephalad and caudal to the lesion of interest. Lesion size measurement under RECISTS 1.1 guidelines, choice of CT slice, and time spent on measurement were captured. Prior to their final measurements, the participants were exposed to the intervention designed to reduce the number of measurements deviating from the median. Chi-square test was performed to identify radiologist-dependent factors associated with the variability. The percent of deviating measurements during T1 and T2 were 20.0% and 23.1%, respectively. There was no statistically significant change in the number of deviating measurements upon the presentation of the intervention despite the decrease in percent from 23.1% to 17.7%. The identified barriers to the intervention include clinical disagreements among radiologists. Specifically, the inter-observer variability was associated with the controversy over the choice of CT image slice (P = 0.045) and selection of start-point, axis, and end-point (P = 0.011). Clinical disagreements rather than random errors were barriers to reducing interobserver variability in CT measurement among experienced radiologists. Future interventions could aim to resolve the disagreement in an interactive approach.

Retrospective comparison of approaches to evaluating inter-observer variability in CT tumour measurements in an academic health centre.

BACKGROUND: A growing number of research studies have reported inter-observer variability in sizes of tumours measured from CT scans. It remains unclear whether the conventional statistical measures correctly evaluate the CT measurement consistency for optimal treatment management and decision-making. We compared and evaluated the existing measures for evaluating inter-observer variability in CT measurement of cancer lesions. METHODS: 13 board-certified radiologists repeatedly reviewed 10 CT image sets of lung lesions and hepatic metastases selected through a randomisation process. A total of 130 measurements under RECIST 1.1 (Response Evaluation Criteria in Solid Tumors) guidelines were collected for the demonstration. Intraclass correlation coefficient (ICC), Bland-Altman plotting and outlier counting methods were selected for the comparison. The each selected measure was used to evaluate three cases with observed, increased and decreased inter-observer variability. RESULTS: The ICC score yielded a weak detection when evaluating different levels of the inter-observer variability among radiologists (increased: 0.912; observed: 0.962; decreased: 0.990). The outlier counting method using Bland-Altman plotting with 2SD yielded no detection at all with its number of outliers unchanging regardless of level of inter-observer variability. Outlier counting based on domain knowledge was more sensitised to different levels of the inter-observer variability compared with the conventional measures (increased: 0.756; observed: 0.923; improved: 1.000). Visualisation of pairwise Bland-Altman bias was also sensitised to the inter-observer variability with its pattern rapidly changing in response to different levels of the inter-observer variability. CONCLUSIONS: Conventional measures may yield weak or no detection when evaluating different levels of the inter-observer variability among radiologists. We observed that the outlier counting based on domain knowledge was sensitised to the inter-observer variability in CT measurement of cancer lesions. Our study demonstrated that, under certain circumstances, the use of standard statistical correlation coefficients may be misleading and result in a sense of false security related to the consistency of measurement for optimal treatment management and decision-making.

Effect of clinical decision rules, patient cost and malpractice information on clinician brain CT image ordering: a randomized controlled trial.

BACKGROUND: The frequency of head computed tomography (CT) imaging for mild head trauma patients has raised safety and cost concerns. Validated clinical decision rules exist in the published literature and on-line sources to guide medical image ordering but are often not used by emergency department (ED) clinicians. Using simulation, we explored whether the presentation of a clinical decision rule (i.e. Canadian CT Head Rule - CCHR), findings from malpractice cases related to clinicians not ordering CT imaging in mild head trauma cases, and estimated patient out-of-pocket cost might influence clinician brain CT ordering. Understanding what type and how information may influence clinical decision making in the ordering advanced medical imaging is important in shaping the optimal design and implementation of related clinical decision support systems. METHODS: Multi-center, double-blinded simulation-based randomized controlled trial. Following standardized clinical vignette presentation, clinicians made an initial imaging decision for the patient. This was followed by additional information on decision support rules, malpractice outcome review, and patient cost; each with opportunity to modify their initial order. The malpractice and cost information differed by assigned group to test the any temporal relationship. The simulation closed with a second vignette and an imaging decision. RESULTS: One hundred sixteen of the 167 participants (66.9%) initially ordered a brain CT scan. After CCHR presentation, the number of clinicians ordering a CT dropped to 76 (45.8%), representing a 21.1% reduction in CT ordering (P = 0.002). This reduction in CT ordering was maintained, in comparison to initial imaging orders, when presented with malpractice review information (p = 0.002) and patient cost information (p = 0.002). About 57% of clinicians changed their order during study, while 43% never modified their imaging order. CONCLUSION: This study suggests that ED clinician brain CT imaging decisions may be influenced by clinical decision support rules, patient out-of-pocket cost information and findings from malpractice case review. TRIAL REGISTRATION: NCT03449862 , February 27, 2018, Retrospectively registered.