Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ImmunoDiff: A Diffusion Model for Immunotherapy Response Prediction in Lung Cancer

Moinak Bhattacharya, Judy Huang, Amna F. Sher, Gagandeep Singh, Chao Chen, Prateek Prasanna

TL;DR

The paper addresses predicting NSCLC immunotherapy response and survival by translating baseline CT to post-treatment CT. It introduces ImmunoDiff, an anatomy- and clinically guided diffusion model that uses lobar and vascular structure priors and a cbi-Adapter to integrate clinical variables. The method comprises two stages: Stage 1 anatomy-guided diffusion pretraining; Stage 2 clinical-variable-guided post-treatment generation and downstream prediction. On an in-house NSCLC cohort treated with ICIs, ImmunoDiff achieves substantial gains in balanced accuracy and c-index, supporting its potential for improving response prediction and guiding treatment.

Abstract

Accurately predicting immunotherapy response in Non-Small Cell Lung Cancer (NSCLC) remains a critical unmet need. Existing radiomics and deep learning-based predictive models rely primarily on pre-treatment imaging to predict categorical response outcomes, limiting their ability to capture the complex morphological and textural transformations induced by immunotherapy. This study introduces ImmunoDiff, an anatomy-aware diffusion model designed to synthesize post-treatment CT scans from baseline imaging while incorporating clinically relevant constraints. The proposed framework integrates anatomical priors, specifically lobar and vascular structures, to enhance fidelity in CT synthesis. Additionally, we introduce a novel cbi-Adapter, a conditioning module that ensures pairwise-consistent multimodal integration of imaging and clinical data embeddings, to refine the generative process. Additionally, a clinical variable conditioning mechanism is introduced, leveraging demographic data, blood-based biomarkers, and PD-L1 expression to refine the generative process. Evaluations on an in-house NSCLC cohort treated with immune checkpoint inhibitors demonstrate a 21.24% improvement in balanced accuracy for response prediction and a 0.03 increase in c-index for survival prediction. Code will be released soon.

ImmunoDiff: A Diffusion Model for Immunotherapy Response Prediction in Lung Cancer

TL;DR

The paper addresses predicting NSCLC immunotherapy response and survival by translating baseline CT to post-treatment CT. It introduces ImmunoDiff, an anatomy- and clinically guided diffusion model that uses lobar and vascular structure priors and a cbi-Adapter to integrate clinical variables. The method comprises two stages: Stage 1 anatomy-guided diffusion pretraining; Stage 2 clinical-variable-guided post-treatment generation and downstream prediction. On an in-house NSCLC cohort treated with ICIs, ImmunoDiff achieves substantial gains in balanced accuracy and c-index, supporting its potential for improving response prediction and guiding treatment.

Abstract

Accurately predicting immunotherapy response in Non-Small Cell Lung Cancer (NSCLC) remains a critical unmet need. Existing radiomics and deep learning-based predictive models rely primarily on pre-treatment imaging to predict categorical response outcomes, limiting their ability to capture the complex morphological and textural transformations induced by immunotherapy. This study introduces ImmunoDiff, an anatomy-aware diffusion model designed to synthesize post-treatment CT scans from baseline imaging while incorporating clinically relevant constraints. The proposed framework integrates anatomical priors, specifically lobar and vascular structures, to enhance fidelity in CT synthesis. Additionally, we introduce a novel cbi-Adapter, a conditioning module that ensures pairwise-consistent multimodal integration of imaging and clinical data embeddings, to refine the generative process. Additionally, a clinical variable conditioning mechanism is introduced, leveraging demographic data, blood-based biomarkers, and PD-L1 expression to refine the generative process. Evaluations on an in-house NSCLC cohort treated with immune checkpoint inhibitors demonstrate a 21.24% improvement in balanced accuracy for response prediction and a 0.03 increase in c-index for survival prediction. Code will be released soon.

Paper Structure

This paper contains 6 sections, 2 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Stage 1: Anatomy-guided training
  4. Stage 2: Clinical variable-guided response prediction
  5. Results
  6. Conclusion

Figures (2)

  • Figure 1: ImmunoDiff pipeline.(A) Vessels and lobes masks are pre-trained in a contrastive manner. (B) A DDPM model is trained on a large cohort of NSCLC patient CT images. (C) A ControlNet is trained with the vessel and lobe as controls. (D) Clinical variables are used as controls in addition with pre-treatment CT images to generate post-treatment CT images. (E) cbi-Adapter architecture. (F) The trained diffusion model is used for immunotherapy response prediction tasks (responder vs. non-responder classification and survival prediction).
  • Figure 2: Qualitative results. (A) CT images generated w/o pre-training and with pre-training but w/o anatomy-guided training. (B) CT image generated using anatomy-guided training. Also shown are the lobes and vessels used as controls. (C) Post-treatment CT generated by ImmunoDiff for responder and non-responder cases. (tumors shown in red box).