Department seminar: Imaging and Radiomics in Oncology November 20th 12-1pm
Professor Gregory Czarnota
Sunnybrook Health Sciences Centre
Imaging and Radiomics in Oncology
Date: Wednesday November 20th 12-1pm
Location: LAS 3033
Lunch bites will be provided.
Previous studies have demonstrated that quantitative ultrasound (QUS) is an effective tool for monitoring breast cancer patients undergoing neoadjuvant chemotherapy (NAC). Here, we demonstrate the clinical utility of pre-treatment QUS texture features from core-margin and machine learning analysis in predicting the response of breast cancer patients to NAC. Using a 6 MHz center frequency clinical ultrasound imaging system, radiofrequency (RF) breast ultrasound data were acquired from 100 locally advanced breast cancer (LABC) patients prior to their NAC treatment. QUS Spectral parameters were computed from regions of interest (ROI) in the tumor core and its margin. Subsequently, employing gray-level co-occurrence matrices (GLCM), textural features were examined as potential predictive indicators. QUS results were compared with the clinical/pathological response of each patient determined at the end of their NAC and recurrence free survival (RFS) after 2 years clinical follow-up. Three machine learning algorithms based on linear discriminant, k-nearest-neighbors, and support vector machine were developed based on estimated parameters and classification results were compared. Using a 6 MHz ultrasound system, radiofrequency (RF) ultrasound data were acquired from 100 patients with biopsy-confirmed locally advanced breast cancer prior to the start of their NAC. Quantitative ultrasound mean parameter intensity and texture features were computed from tumour core and its margin, and were compared to clinical/pathological response and 5-year recurrence-free survival (RFS) of patients. A multi-parametric QUS model in conjunction with an k-nearest neighbor classifier could predict patient response with 86 % sensitivity, 90% specificity, 90% accuracy, and a 0.88 area under the receiver operating characteristic curve (AUC). The model predicted patient RFS with 94% sensitivity, 77% specificity, 91% accuracy, and a 0.88 AUC. The application of machine learning for classifying patient response based on their QUS features performs well in terms of predicting responders versus non-responders. The findings here provide a framework for developing personalized pre-treatment chemotherapy selection for patients, potentially resulting in improved patient prognosis.
Dr. Gregory Czarnota is a Radiation Oncologist and the Chief of the Radiation Oncology Department at the Odette Cancer Centre. He is Professor in the Departments of Radiation Oncology and Medical Biophysics at the University of Toronto and Senior Scientist and Director of the Odette Cancer Research Program at Sunnybrook Research Institute. He is also at the Sunnybrook Research Institute. Dr. Czarnota’s research is focused on developing methods to detect apoptosis and other forms of cell death in response to cancer therapy, and his work has revolutionized the detection of cell death by ultrasound imaging.