Joseph Lo
Overview:
My research uses computer vision and machine learning to improve medical imaging, focusing on breast and CT imaging. There are three specific projects:
(1) We design deep learning models to diagnose breast cancer from mammograms. We perform single-shot lesion detection, multi-task segmentation/classification, and image synthesis. Our goal is to improve radiologist diagnostic performance and empower patients to make personalized treatment decisions. This work is funded by NIH, Dept of Defense, Cancer Research UK, and other agencies.
(2) We create "digital twin" anatomical models that are based on actual patient data and thus contain highly realistic anatomy. With customized 3D printing, these virtual phantoms can also be rendered into physical form to be scanned on actual imaging devices, which allows us to assess image quality in new ways that are clinically relevant.
(3) We are building a computer-aided triage platform to classify multiple diseases across multiple organs in chest-abdomen-pelvis CT scans. Our hospital-scale data sets have hundreds of thousands of patients. This work includes natural language processing to analyze radiology reports as well as deep learning models for organ segmentation and disease classification.
Positions:
Professor in Radiology
Professor in the Department of Electrical and Computer Engineering
Member of the Duke Cancer Institute
Education:
B.S.E.E. 1988
Ph.D. 1993
Research Associate, Radiology
Grants:
Predicting Breast Cancer With Ultrasound and Mammography
Improved Diagnosis of Breast Microcalcification Clusters
Accurate Models for Predicting Radiation-Induced Injury
Computer Aid for the Decision to Biopsy Breast Lesions
Computer Aid for the Decision to Biopsy Breast Lesions
Publications:
Utility of a Rule-Based Algorithm in the Assessment of Standardized Reporting in PI-RADS.
Anomaly Detection of Calcifications in Mammography Based on 11,000 Negative Cases.
Multi-label annotation of text reports from computed tomography of the chest, abdomen, and pelvis using deep learning.
Prediction of Upstaging in Ductal Carcinoma in Situ Based on Mammographic Radiomic Features.
Technical note: Controlling the attenuation of 3D-printed physical phantoms for computed tomography with a single material.
Research Areas:
