Deep Semisupervised Transfer Learning for Fully Automated Whole-Body Tumor Quantification and Prognosis of Cancer on PET/CT.

in Journal of nuclear medicine : official publication, Society of Nuclear Medicine by Kevin H Leung, Steven P Rowe, Moe S Sadaghiani, Jeffrey P Leal, Esther Mena, Peter L Choyke, Yong Du, Martin G Pomper

TLDR

  • The study developed a deep learning approach to automatically detect and characterize cancer from PET/CT scans. The approach learned to segment tumors and make predictions about patient outcomes. The study found that the approach worked well for several types of cancer, including prostate, lung, and breast cancer. The results could help doctors make more accurate diagnoses and better treatment plans for patients with cancer.

Abstract

Automatic detection and characterization of cancer are important clinical needs to optimize early treatment. We developed a deep, semisupervised transfer learning approach for fully automated, whole-body tumor segmentation and prognosis on PET/CT.This retrospective study consisted of 611F-FDG PET/CT scans of patients with lung cancer, melanoma, lymphoma, head and neck cancer, and breast cancer and 408 prostate-specific membrane antigen (PSMA) PET/CT scans of patients with prostate cancer. The approach had a nnU-net backbone and learned the segmentation task onF-FDG and PSMA PET/CT images using limited annotations and radiomics analysis. True-positive rate and Dice similarity coefficient were assessed to evaluate segmentation performance. Prognostic models were developed using imaging measures extracted from predicted segmentations to perform risk stratification of prostate cancer based on follow-up prostate-specific antigen levels, survival estimation of head and neck cancer by the Kaplan-Meier method and Cox regression analysis, and pathologic complete response prediction of breast cancer after neoadjuvant chemotherapy. Overall accuracy and area under the receiver-operating-characteristic (AUC) curve were assessed.Our approach yielded median true-positive rates of 0.75, 0.85, 0.87, and 0.75 and median Dice similarity coefficients of 0.81, 0.76, 0.83, and 0.73 for patients with lung cancer, melanoma, lymphoma, and prostate cancer, respectively, on the tumor segmentation task. The risk model for prostate cancer yielded an overall accuracy of 0.83 and an AUC of 0.86. Patients classified as low- to intermediate- and high-risk had mean follow-up prostate-specific antigen levels of 18.61 and 727.46 ng/mL, respectively (< 0.05). The risk score for head and neck cancer was significantly associated with overall survival by univariable and multivariable Cox regression analyses (< 0.05). Predictive models for breast cancer predicted pathologic complete response using only pretherapy imaging measures and both pre- and posttherapy measures with accuracies of 0.72 and 0.84 and AUCs of 0.72 and 0.76, respectively.The proposed approach demonstrated accurate tumor segmentation and prognosis in patients across 6 cancer types onF-FDG and PSMA PET/CT scans.

Overview

  • The study aimed to develop a deep, semisupervised transfer learning approach for fully automated, whole-body tumor segmentation and prognosis on PET/CT scans of patients with various types of cancer. The approach used limited annotations and radiomics analysis to learn the segmentation task on F-FDG and PSMA PET/CT images. The study evaluated the performance of the approach on F-FDG and PSMA PET/CT scans of patients with lung cancer, melanoma, lymphoma, head and neck cancer, breast cancer, and prostate cancer. The primary objective of the study was to demonstrate accurate tumor segmentation and prognosis in patients across 6 cancer types on F-FDG and PSMA PET/CT scans.

Comparative Analysis & Findings

  • The study compared the outcomes observed under different experimental conditions or interventions detailed in the study. The results showed that the proposed approach yielded median true-positive rates of 0.75, 0.85, 0.87, and 0.75 and median Dice similarity coefficients of 0.81, 0.76, 0.83, and 0.73 for patients with lung cancer, melanoma, lymphoma, and prostate cancer, respectively, on the tumor segmentation task. The risk model for prostate cancer yielded an overall accuracy of 0.83 and an AUC of 0.86. Patients classified as low- to intermediate- and high-risk had mean follow-up prostate-specific antigen levels of 18.61 and 727.46 ng/mL, respectively (< 0.05). The risk score for head and neck cancer was significantly associated with overall survival by univariable and multivariable Cox regression analyses (< 0.05). Predictive models for breast cancer predicted pathologic complete response using only pretherapy imaging measures and both pre- and posttherapy measures with accuracies of 0.72 and 0.84 and AUCs of 0.72 and 0.76, respectively. The study identified significant differences in the results between these conditions, with the proposed approach demonstrating accurate tumor segmentation and prognosis in patients across 6 cancer types on F-FDG and PSMA PET/CT scans.

Implications and Future Directions

  • The study's findings have significant implications for the field of research or clinical practice. The proposed approach demonstrated accurate tumor segmentation and prognosis in patients across 6 cancer types on F-FDG and PSMA PET/CT scans, which could improve early treatment and risk stratification of cancer patients. The study identified limitations, such as the need for more data and validation in larger cohorts. Future research directions could include exploring the use of the proposed approach in other types of cancer, improving the accuracy of the segmentation task, and integrating the approach with other clinical tools and workflows.
Read Full Article