Automatic 3D Segmentation and Longitudinal Volumetric Measurement of Breast Lesions For Automated Breast Ultrasound Images
Monday, April 8, 2024
3:04pm – 3:11pm
Location: 412
Authors: Andrew Leynes, iSono Health, Inc. Shadi Saberi, iSono Health, Inc.
The use of automated breast ultrasound (ABUS) is on the rise primarily driven by the shortage of skilled sonographers. One of the key advantages of ABUS is that it eliminates the limitations associated with operator dependence and the labor-intensive nature of handheld 2D ultrasound. However, the utilization of 3D breast ultrasound imposes a heavier workload on radiologists and lengthens the time required for image interpretation and lesion annotation and measurement due to the large number of volumetric slices. We developed an AI-based 3D segmentation model to automatically measure the volume of a detected breast lesion precisely and reproducibly. We have demonstrated that our model evaluated on simulated longitudinal data can reliably detect clinically significant volumetric changes in breast lesions.