Impact of a Categorical AI System for Digital Breast Tomosynthesis on Breast Cancer Interpretation by Both General Radiologists and Breast Imaging Specialists.

Purpose To evaluate performance improvements of general radiologists and breast imaging specialists when interpreting a set of diverse digital breast tomosynthesis (DBT) examinations with the aid of a custom-built categorical artificial intelligence (AI) system. Materials and Methods A fully balanced multireader, multicase reader study was conducted to compare the performance of 18 radiologists (nine general radiologists and nine breast imaging specialists) reading 240 retrospectively collected screening DBT mammograms (mean patient age, 59.8 years ± 11.3 [SD]; 100% women), acquired between August 2016 and March 2019, with and without the aid of a custom-built categorical AI system. The area under the receiver operating characteristic curve (AUC), sensitivity, and specificity across general radiologists and breast imaging specialists reading with versus without AI were assessed. Reader performance was also analyzed as a function of breast cancer characteristics and patient subgroups. Results Every radiologist demonstrated improved interpretation performance when reading with versus without AI, with an average AUC of 0.93 versus 0.87, demonstrating a difference in AUC of 0.06 (95% CI: 0.04, 0.08; P < .001). Improvement in AUC was observed for both general radiologists (difference of 0.08; P < .001) and breast imaging specialists (difference of 0.04; P < .001) and across all cancer characteristics (lesion type, lesion size, and pathology) and patient subgroups (race and ethnicity, age, and breast density) examined. Conclusion A categorical AI system helped improve overall radiologist interpretation performance of DBT screening mammograms for both general radiologists and breast imaging specialists and across various patient subgroups and breast cancer characteristics. Keywords: Computer-aided Diagnosis, Screening Mammography, Digital Breast Tomosynthesis, Breast Cancer, Screening, Convolutional Neural Network (CNN), Artificial Intelligence Supplemental material is available for this article. © RSNA, 2024.

Robust breast cancer detection in mammography and digital breast tomosynthesis using an annotation-efficient deep learning approach.

Breast cancer remains a global challenge, causing over 600,000 deaths in 2018 (ref. 1). To achieve earlier cancer detection, health organizations worldwide recommend screening mammography, which is estimated to decrease breast cancer mortality by 20-40% (refs. 2,3). Despite the clear value of screening mammography, significant false positive and false negative rates along with non-uniformities in expert reader availability leave opportunities for improving quality and access4,5. To address these limitations, there has been much recent interest in applying deep learning to mammography6-18, and these efforts have highlighted two key difficulties: obtaining large amounts of annotated training data and ensuring generalization across populations, acquisition equipment and modalities. Here we present an annotation-efficient deep learning approach that (1) achieves state-of-the-art performance in mammogram classification, (2) successfully extends to digital breast tomosynthesis (DBT; '3D mammography'), (3) detects cancers in clinically negative prior mammograms of patients with cancer, (4) generalizes well to a population with low screening rates and (5) outperforms five out of five full-time breast-imaging specialists with an average increase in sensitivity of 14%. By creating new 'maximum suspicion projection' (MSP) images from DBT data, our progressively trained, multiple-instance learning approach effectively trains on DBT exams using only breast-level labels while maintaining localization-based interpretability. Altogether, our results demonstrate promise towards software that can improve the accuracy of and access to screening mammography worldwide.

Learning disease relationships from clinical drug trials.

OBJECTIVE: Our objective is to test the limits of the assumption that better learning from data in medicine requires more granular data. We hypothesize that clinical trial metadata contains latent scientific, clinical, and regulatory expert knowledge that can be accessed to draw conclusions about the underlying biology of diseases. We seek to demonstrate that this latent information can be uncovered from the whole body of clinical trials. MATERIALS AND METHODS: We extract free-text metadata from 93 654 clinical drug trials and introduce a representation that allows us to compare different trials. We then construct a network of diseases using only the trial metadata. We view each trial as the summation of expert knowledge of biological mechanisms and medical evidence linking a disease to a drug believed to modulate the pathways of that disease. Our network representation allows us to visualize disease relationships based on this underlying information. RESULTS: Our disease network shows surprising agreement with another disease network based on genetic data and on the Medical Subject Headings (MeSH) taxonomy, yet also contains unique disease similarities. DISCUSSION AND CONCLUSION: The agreement of our results with other sources indicates that our premise regarding latent expert knowledge holds. The disease relationships unique to our network may be used to generate hypotheses for future biological and clinical research as well as drug repurposing and design. Our results provide an example of using experimental data on humans to generate biologically useful information and point to a set of new and promising strategies to link clinical outcomes data back to biological research.

Microfluidic reflow pumps.

A new microfluidic pump, termed a reflow pump, is designed to operate with a sub-µl sample volume and transport it back and forth between two pneumatically actuated reservoirs through a flow channel typically containing one or more sensor surfaces. The ultimate motivation is to efficiently use the small sample volume in conjunction with convection to maximize analyte flux to the sensor surface(s) in order to minimize sensor response time. In this paper, we focus on the operational properties of the pumps themselves (rather than the sensor surfaces), and demonstrate both two-layer and three-layer polydimethylsiloxane reflow pumps. For the three-layer pump, we examine the effects of reservoir actuation pressure and actuation period, and demonstrate average volumetric flow rates as high as 500 µl/min. We also show that the two-layer design can pump up to 93% of the sample volume during each half period and demonstrate integration of a reflow pump with a single-chip microcantilever array to measure maximum flow rate.

Unilateral hypoglossal neurapraxia following endotracheal intubation for total shoulder arthroplasty.

A case is described of postoperative right hypoglossal neurapraxia after general anesthesia and interscalene block with endotracheal intubation for left total shoulder arthroplasty. Postoperative hypoglossal neurapraxia has been reported in cases, yet it remains a rare complication of anesthesia-related interventions. In this case report, postulated causes of hypoglossal neurapraxia are presented. A review of the literature pertaining to anesthesia-related causes of hypoglossal nerve injury is included. Anesthesia providers should be aware of the course of cranial nerve XII as it relates to the position of the head and neck and use of airway instrumentation. In suspected cases of hypoglossal neurapraxia, conservative therapeutic interventions may be warranted.

Three-dimensional intracellular optical coherence phase imaging.

Quantitative phase imaging has many applications for label-free studies of the nanoscale structure and dynamics of cells and tissues. It has been demonstrated that optical coherence phase microscopy (OCPM) can provide quantitative phase information with very high sensitivity. The excellent phase stability of OCPM is obtained by use of a reflection from the microscope cover glass as a local reference field. For detailed intracellular studies a large numerical aperture (N.A.) objective is needed in order to obtain the required resolution. Unfortunately, this also means that the depth of field becomes too small to obtain sufficient power from the cover glass when the beam is focused into the sample. To address this issue, we designed a setup with a dual-beam sample arm. One beam with a large diameter (filling the 1.2 N.A. water immersion objective) enabled high-resolution imaging. A second beam with a small diameter (underfilling the same objective) had a larger depth of field and could detect the cover glass used as a local phase reference. The phase stability of the setup was quantified by monitoring the front and back of a cover glass. The standard deviation of the phase difference was 0.021 rad, corresponding to an optical path displacement of 0.9 nm. The lateral and axial dimensions of the confocal point spread function were 0.42 and 0.84 µm, respectively. This makes our dual-beam setup ideal for three-dimensional intracellular phase imaging.

Distilling clinically interpretable information from data collected on next-generation wearable sensors.

Medical electronic systems are generating ever larger data sets from a variety of sensors and devices. Such systems are also being packaged in wearable designs for easy and broad use. The large volume of data and the constraints of low-power, extended-duration, and wireless monitoring impose the need for on-chip processing to distill clinically relevant information from the raw data. The higher-level information, rather than the raw data, is what needs to be transmitted. We present one example of information processing for continuous, high-sampling-rate data collected from wearable and portable devices. A wearable cardiac and motion monitor designed by colleagues at MIT simultaneously records electrocardiogram (ECG) and 3-axis acceleration to onboard memory, in an ambulatory setting. The acceleration data is used to generate a continuous estimate of physical activity. Additionally, we use a Portapres continuous blood pressure monitor to concurrently record the arterial blood pressure (ABP) waveform. To help reduce noise, which is an increased challenge in ambulatory monitoring, we use both the ECG and ABP waveforms to generate a robust measure of heart rate from noisy data. We also generate an overall signal abnormality index to aid in the interpretation of the results. Two important cardiovascular quantities, namely cardiac output (CO) and total peripheral resistance (TPR), are then derived from this data over a sequence of physical activities. CO and TPR can be estimated (to within a scale factor) from heart rate, pulse pressure and mean arterial blood pressure, which in turn are directly obtained from the ECG and ABP signals. Data was collected on 10 healthy subjects. The derived quantities vary in a manner that is consistent with known physiology. Further work remains to correlate these values with the cardiac health state.