Deep Learning and Radiomics for Medical Imaging in Oncology and Radiotherapy
This project reads clinical MRI with deep learning and radiomics — segmenting the prostate and its zones, classifying suspicious lesions, and automating the contours that adaptive radiotherapy depends on.
Olmo Zavala-Romero Yifan Wang Radka StoyanovaAdrian L. BretoDeukwoo KwonAlan PollackJohn C. Ford
Goals
Develop and validate machine learning and deep learning methods that extract quantitative information from clinical MRI and fMRI to support cancer diagnosis, risk stratification, and adaptive radiotherapy. This research line spans prostate multiparametric MRI (mpMRI)—from lesion classification and zonal segmentation to integration with clinical-genomic risk models—and extends to MRI-guided radiotherapy workflows for cervical, pancreatic, and glioblastoma patients, plus resting-state fMRI studies of brain network connectivity.
Methods
We apply radiomics and machine learning to multiparametric prostate MRI for clinically significant lesion classification (PROSTATEx challenge framework) and multi-vendor prostate and peripheral zone segmentation using 3D multistream convolutional neural networks. For adaptive radiotherapy, we develop automated pipelines for gross tumor volume (GTV) and organ-at-risk (OAR) delineation on daily and per-fraction MRI, AI-based abdominal anatomy segmentation in pancreatic adaptive RT, multi-reference tissue normalization on T2-weighted MRI, and automated volume delineation in glioblastoma. Complementary resting-state fMRI connectivity analysis characterizes network changes in nicotine-dependent patients.
Key findings
Published across eight journal and conference papers (2016–2023), this line delivered vendor-robust prostate and peripheral-zone segmentation on 550+ MRIs, radiomics classifiers for clinically significant lesions linked to NCCN and Decipher risk stratification, and deep learning pipelines that automate contouring in adaptive radiotherapy for cervical, pancreatic, and glioblastoma patients. The work was carried out in collaboration with the University of Miami radiation oncology and radiomics groups and established foundational expertise in volumetric MRI analysis now extended to other biomedical imaging at the lab.
Publications from this project
- Functional Connectivity Analysis of Resting-State fMRI Networks in Nicotine Dependent Patients 2016
- Classification of Suspicious Lesions on Prostate Multiparametric MRI Using Machine Learning 2018
- A Deep Learning Pipeline for per-Fraction Automatic Segmentation of GTV and OAR in Cervical Cancer 2019
- Automatic Segmentation of Abdominal Anatomy by Artificial Intelligence in Adaptive Radiotherapy of Pancreatic Cancer 2019
- Segmentation of Prostate and Prostate Zones Using Deep Learning - A Multi-MRI Vendor Analysis 2020
- Deep Learning Approach for Multi-Reference Tissue Normalization on T2-Weighted MRI in Prostate Cancer Radiotherapy 2020
- A Deep Learning Approach for Automated Volume Delineation on Daily MRI Scans in Glioblastoma Patients 2021
- Clinical-Genomic Risk Group Classification of Suspicious Lesions on Prostate Multiparametric-MRI 2023