Segmentation of Lesions in Whole-Body PET/CT Using Deep Learning

Presenter: Tyler Wellman, Ph.D. | Associate Director Oncology, Invicro

This talk reviews some initial successes in applying deep learning to segment lesions in 3D whole-body PET/CT data. The current technique is based on a slice-and-fuse approach, in which independent processing of 2D transverse slices is followed by reconstruction into a 3D volume. The technique considers dual-channel PET/CT inputs, and current/future work is focused on extending the method to accept 3D input data.

During this webinar, the presenter will review:

  1. How deep learning shows impressive performance in segmentation of lesions on PET/CT, with high accuracy and good sensitivity and specificity for lesion-derived uptake on PET.
  2. The use of both modalities as input channels allows for excellent results, even when the problem is approached using 2D slicing of 3D data.

Pre-Register for On-Demand Access:



Associate Director Oncology, Invicro

Dr. Wellman is a clinical imaging scientist currently serving as Associate Director, Oncology at Invicro. He received his Ph.D. from Boston University with a focus on PET imaging of the lungs to study ventilator-induced lung injury. Dr. Wellman has been with Invicro since 2014, working primarily on the analysis of PET/SPECT/CT/MRI images for both preclinical and clinical trials across multiple disease areas. Most recently, his work has centered on quantification of imaging biomarkers for oncology trials, including the use of advanced analytics and machine learning to derive added value from multi-modal imaging data.