Automated fetal anatomy identification to improve quality of obstetrical ultrasound

Abnormal fetal growth is a leading cause of perinatal mortality and morbidity in both developed and developing countries. Fetal growth is primarily assessed by ultrasound due to its low risk, low cost, and wide availability. These studies are cognitively intense for a reporting radiologist and errors can have significant long-term implications. In this study, we will train a novel deep learning algorithm to detect fetal body parts to improve quality in ultrasound reporting according to the Canadian Association of Radiology standard. We first develop a novel, large dataset for classifying obstetrical ultrasound images by body part and grading quality. Next, we develop a deep learning algorithm to detect fetal body parts present in 2D ultrasound image and grade quality. Finally, we will integrate this tool and evaluate impact on radiologist workflow.

Faculty Supervisor:

Dafna Sussman

Student:

Partner:

Trillium Health Partners

Discipline:

Engineering

Sector:

Health and Related Sciences & Technology

University:

Toronto Metropolitan University

Program: