• Center on Health Equity & Access
  • Clinical
  • Health Care Cost
  • Health Care Delivery
  • Insurance
  • Policy
  • Technology
  • Value-Based Care

Deep Learning Model Predicts Response to Combination Therapy Prior to Treatment in HCC

News
Article

Researchers developed a web server that can help predict therapeutic responses to combination lenvatinib and immune checkpoint inhibitors in patients with unresectable hepatocellular carcinoma (HCC).

A multiphase model integrating multiphase CT images with deep learning–generated heatmaps predicted responses to treatment with combination lenvatinib and immune checkpoint inhibitors in patients with unresectable hepatocellular carcinoma (HCC), according to a study published in Computational and Structural Biotechnology Journal. The tool showed excellent performance and could help individualize treatment for patients with unresectable HCC, the study authors concluded.

Combining lenvatinib and immune checkpoint inhibitors is a potentially effective treatment approach for unresectable HCC, but only a subset of patients experience responses to this strategy. Therefore, there is a need for methods to determine which patients are most likely to benefit from lenvatinib plus immune checkpoint inhibitors.

Study authors developed the interpretive deep learning model using multiphase CT images to preemptively determine whether patients would respond to lenvatinib plus immune checkpoint inhibitors based on RECIST v1.1 criteria. The model’s performance was trained and evaluated in a retrospective cohort of 120 patients with unresectable HCC, with patients divided into a training cohort (n = 72) and a validation cohort (n = 48). The overall response rates based on RECIST 1.1 criteria in each cohort were 27.8% and 27.1%, respectively.

In addition to the model’s performance, the authors assessed the impact of each CT phase on prediction accuracy. The dataset comprised a total of 10,815 CT images.

CT scan machine | Image credit: zlikovec - stock.adobe.com

CT scan machine | Image credit: zlikovec - stock.adobe.com

Compared with biphase and uniphase models, the multiphase model that included plain scan, arterial, and portal phases showed superior performance with an area under the curve (AUC) of 0.802 (95% CI, 0.780-0.824). A biphase model based on plain scan and portal phases had an AUC of 0.760, a model based on arterial and portal phases achieved an AUC of 0.740, and a model based on portal and plain scan phases reached an AUC of 0.719. Regarding phase importance, portal phase images significantly enhanced the model’s accuracy and produced better metrics compared with models based on plain scan or arterial phases alone.

The researchers also created heatmaps in the analysis to highlight the regions most crucial to accurate prediction of treatment response, noting 6 clinical features: necrosis, vasculature, tumor heterogeneity, tumor and peritumoral region, intrahepatic multifocal tumors, and tumor regions surrounding transarterial chemoembolization iodide.

“These results underscore deep learning's ability to capture the complex spatial variability of tumors and the impact of the tumor microenvironment on immune sensitization,” the authors wrote. “By providing such detailed insights, deep learning enhances its clinical relevance, helping clinicians identify key areas of interest and thereby facilitating more informed clinical decisions.”

The authors noted several study limitations, including the need for enhancing the accuracy and sensitivity levels of the model to meet clinical standards. Data were also sourced from a single institution, potentially limiting the generalizability of the findings. Finally, the methods used in the study cannot replicate real-world factors and equipment settings.

Despite the limitations, the authors concluded that the findings represent a significant advance in prediction of treatment responses in patients undergoing chemoimmunotherapy for HCC utilizing deep learning. The user-friendly heatmap incorporated into the workflow may also help guide clinical practice, they added. The authors also developed a web server platform to make the model accessible and easy to use.

Reference

Liao NQ, Deng ZJ, Wei W. Deep learning of pretreatment multiphase CT images for predicting response to lenvatinib and immune checkpoint inhibitors in unresectable hepatocellular carcinoma. Comput Struct Biotechnol J. Published online April 3, 2024. doi:10.1016/j.csbj.2024.04.001

Related Videos
Sarah Manes
Amit Singal, MD, UT Southwestern Medical Center
Amit Singal, MD
Michael Morse, MD, Duke Cancer Center
Michael Morse, MD, Duke University Cancer Center
Amit Singal, MD, UT Southwestern Medical Center
Dr Michael Morse, Duke University
Michael Morse, MD, FACP, MHS, Duke Cancer Center
Related Content
© 2024 MJH Life Sciences
AJMC®
All rights reserved.