Search

Steatosis is the most common cause of chronic liver disease worldwide. Ultrasound helps diagnose steatosis due to its wide availability, non-invasiveness, and safety. The hepatorenal index (HRI) is a clinically used biomarker for diagnosing steatosis. To study the automated calculation of HRI from ultrasound images, we identified a retrospective cohort undergoing abdominal ultrasound with 1320 images, one image per patient. The patients were divided into training (n=1166) and test (n=154) sets. The test set consists of 48 patients without steatosis (S0) and 106 patients with steatosis (S1-S3). An automated method was proposed to calculate HRI using four steps (a) segment the liver and the renal cortex, (b) remove pixels associated with ducts, vessels, masses, cysts, and shadowing artifacts, (c) place two regions of interest (ROIs), one each in the liver and the renal cortex, and (d) calculate the HRI as the ratio of the mean pixel value of the liver ROI to the renal cortex ROI. We explored several sets of rules for ROI placement. The method minimizing the Euclidean distance between the two ROIs, the best performer, had an area under the ROC (AUROC) curve of 0.87 compared to physicians with an AUROC of 0.82 (p=0.159). In addition, the Euclidean distance constraint showed no dependence on depth for ROI placement. The proposed automated method can be translated into clinical care without modifying existing imaging protocols to optimize the diagnostic capability of ultrasound for hepatic steatosis and assist with triaging patients.

Endometrial thickness is closely related to gyneco-logical function and is an important biomarker in transvaginal ultrasound (TVUS) examinations for assessing female reproductive health. Manual measurement is time-consuming and subject to high inter- and intra- observer variability. In this paper, we present a fully automated endometrial thickness measurement method using deep learning. Our pipeline consists of: 1) endometrium segmentation using a VGG-based U-Net, and 2) endometrial thickness estimation using medial axis transformation. We conducted experimental studies on 137 2D TVUS cases (74/63 secretory phase/proliferative phase). On a test set of 27 cases/277 images, the segmentation Dice score is 0.83. For thickness measurement, we achieved mean absolute error of 1.23/1.38 mm and root mean squared error of 1.79/1.85 mm on two different test sets. The results are considered well within the clinically acceptable range of ±2 mm. Furthermore, our phase-stratified analysis shows that the measurement variance from the secretory phase is higher than that from the proliferative phase, largely due to the high variability of the endometrium appearance in the secretory phase. Future work will extend our current algorithm toward different clinical outcomes for a broader spectrum of clinical applications.

Stereotactic radiosurgery (SRS), which delivers high doses of irradiation in a single or few shots to small targets, has been a standard of care for brain metastases. While very effective, SRS currently requires manually intensive delineation of tumors. In this work, we present a deep learning approach for automated detection and segmentation of brain metastases using multimodal imaging and ensemble neural networks. In order to address small and multiple brain metastases, we further propose a volume-aware Dice loss which optimizes model performance using the information of lesion size. This work surpasses current benchmark levels and demonstrates a reliable AI-assisted system for SRS treatment planning for multiple brain metastases.

Background: There is an epidemic of retinopathy of prematurity (ROP) in low- and middle-income countries (LMIC) in part due to heterogeneity in adherence to best practices for oxygen monitoring in neonatal care units (NCU). The purpose of this study was to determine whether a deep learning based ROP severity score obtained from fundus images in a telemedicine program could be applied at the population level to compare epidemiologic differences between NCUs in South India. Methods: Fundus photos were obtained from at-risk infants at participating hospitals in the Aravind Eye Hospital ROP telemedicine system. An ROP severity score from 1-9 was calculated for each eye examination using methods previously published. Population (NCU) level ROP severity was compared across hospitals using multivariate linear regression adjusting for demographic risk factors for ROP. Masked ROP technicians assessed adherence to best practices using a quality score (1-5) for oxygen monitoring in each NCU. Findings: Images from initial screening exams of 128 infants at 11 hospitals were analyzed. Mean ROP severity was significantly different between hospitals (P=0.003). In multivariate regression controlling for birth weight, gestational age, and postnatal age, a higher NCU quality score was associated with lower ROP severity (P=0.001). Of note, the NCU with the lowest quality score had the highest ROP severity score (1.6 points above the mean, P=0.001), while the NCU with the highest quality score had the lowest ROP severity score (1.2 points below the mean, P=0.001). Interpretation: We identified variability in ROP severity between NCUs that correlated with best practices for oxygen monitoring. These findings suggest the potential utility of using image-based deep learning not just for patient-level diagnosis, but also for population-level epidemiologic surveillance of primary prevention practices, and may have broader applicability for a variety of medical conditions beyond ROP. Funding Statement: This project was supported by grants R01EY19474, K12EY027720, and P30EY10572 from the National Institutes of Health (Bethesda, MD), by grants SCH1622679, SCH-1622542, and SCH-1622536 from the National Science Foundation (Arlington, VA), and by unrestricted departmental funding and a Career Development Award (JPC) from Research to Prevent Blindness (New York, NY). Declaration of Interests: Michael F. Chiang is an unpaid member of the Scientific Advisory Board for Clarity Medical Systems (Pleasanton, CA), a Consultant for Novartis (Basel, Switzerland), and an initial member of Inteleretina (Honolulu, HI). R. V. Paul Chan is on the Scientific Advisory Board for Visunex Medical Systems (Pleasanton, CA), and a Consultant for Genentech (South San Francisco, CA). Michael Chiang and J. Peter Campbell receive research support from Genentech. J. Peter Campbell, James M. Brown, Susan Ostmo, R.V.P. Chan, Jayaskhree Kalpathy-Cramer, and Michael F. Chiang have a patent application pending on the described technology. All other authors have no conflicts of interest. Ethics Approval Statement: This study was conducted in accordance with Health Insurance Portability and Accountability Act (HIPAA) guidelines, and Institutional Review Board (IRB) approval was obtained at both Oregon Health & Science University and the Aravind Eye Hospital.

RationaleMultiple clinical trials support the effectiveness of cardiac resynchronization therapy (CRT); however, optimal patient selection remains challenging due to substantial treatment heterogeneity among patients who meet the clinical practice guidelines.ObjectiveTo apply machine learning to create an algorithm that predicts CRT outcome using electronic health record (EHR) data avaible before the procedure.Methods and resultsWe applied machine learning and natural language processing to the EHR of 990 patients who received CRT at two academic hospitals between 2004-2015. The primary outcome was reduced CRT benefit, defined as ConclusionsA machine learning model that leveraged readily available EHR data and clinical notes identified a subset of CRT patients who may not benefit from CRT before the procedure.