Thursday, November 30 | 1:40 p.m.-1:50 p.m. | R6-SSNMMI08-2 | Room S403A
An AI model based on F-18 DCFPyL PSMA-PET/CT images will be presented here that shows promise for predicting treatment response in patients with metastatic castration-resistant prostate cancer (mCRPC).
Despite the widespread adoption of prostate-specific membrane antigen (PSMA)-PET/CT imaging for prostate cancer, prognostication of patient outcomes remains challenging. The goal of this work was to evaluate the feasibility of a deep-learning model to identify patients at risk for progression based on baseline PSMA-PET, noted Andrew Voter, MD, of Johns Hopkins Medicine in Baltimore, MD, who will present the study.
The researchers used PSMA-PET/CT scans of 128 patients (age 71.7 ± 8.4 years, prostate-specific antigen [PSA] 19.4 ± 42.7) with metastatic castration-resistant prostate cancer (mCRPC) to train a multimodal fusion 3D convolutional neural network (CNN). Input included a total of 1,624 PSMA-avid foci suspicious for malignancy, with lesions classified as either progressive or nonprogressive based on increased radiotracer avidity or a greater than 2-mm size increase.
The model was validated using an independent test set of 105 lesions from 19 patients (age 72.2 ± 7.1 years, PSA 18.5±16.1). The PET/CT model achieved an accuracy of 82%, an area under the curve of 0.61, and an F1-score of 90% in classifying lesions as progressive versus non-progression.
In addition, the algorithm was able to identify a high-risk population in the testing set with significantly reduced median survival (26.1 months) relative to the remaining population (57.4 months, p = 0.04), according to the findings.
“Further studies are warranted to determine the clinical implications of our deep learning model,” Voter noted.
Attend this session to learn the details.
