A Deep Learning Pathomics Platform May Help Predict Response to Immunotherapy in Lung Cancer Patients

SAN DIEGO – A biology-guided artificial intelligence model applied to routine pathology slides accurately predicted outcomes and response to immunotherapy in patients with metastatic non-small cell lung cancer (NSCLC), according to a study presented at the American Association for Cancer Research (AACR) Annual Meeting 2026, held April 17-22.

“Immunotherapy has transformed cancer treatment, but only a subset of patients benefit from it, and predicting who will respond remains challenging,” said presenter Rukhmini Bandyopadhyay, PhD, a postdoctoral fellow at The University of Texas (UT) MD Anderson Cancer Center. “While routine pathology slides contain rich information about the tumor and its surrounding environment, the large amount of complex data can be difficult for human experts to fully quantify.”

Pathomics is an emerging field that applies computational and machine learning methods for high-throughput analysis of digital pathology images to extract and analyze large-scale data related to cell and tissue architecture that can be linked to disease outcomes.

Bandyopadhyay and colleagues developed a pathomics framework based on a deep learning survival prediction model called Pathology-driven Immunotherapy Optimization or Path-IO, which can analyze pathology slide images and study patterns across the tissue to help identify patients who are most likely to benefit from immunotherapy.

“The core idea was to identify specific features, known as niches, within the tumor microenvironment to understand how tumors and the surrounding tissues are organized,” said Bandyopadhyay, adding that the model then combines this information with available imaging and clinical data to estimate whether a patient may have a higher or lower risk of poor outcomes from immunotherapy.

The researchers tested the platform in a study that included 797 immune checkpoint inhibitor-treated NSCLC patients from UT MD Anderson, with external validation in 280 additional patients from Mayo Clinic, Gustave Roussy, and the phase III Lung-MAP S1400I trial in which immunotherapy-naïve patients with lung squamous cell carcinoma (a subtype of NSCLC) were treated with immune checkpoint inhibitors.

Study results showed that the model could reliably stratify patients into higher and lower risk groups with significantly different outcomes. In the UT MD Anderson cohort, patients in the high‑risk group had more than double the risk of death or disease progression compared with patients in the low‑risk group. Comparable results were obtained in the validation datasets.

Model performance was evaluated using the concordance index (C-index), which measures how well each biomarker distinguishes between patients with different outcomes, explained Bandyopadhyay. Notably, Path-IO consistently outperformed PD-L1, the U.S. Food and Drug Administration-validated standard-of-care biomarker for guiding immunotherapy use in NSCLC patients, across both discovery and test cohorts. “PD-L1 alone showed limited prognostic performance, with C-indices of 0.58 for overall survival (OS) and 0.57 for progression-free survival (PFS) in the discovery cohort, declining to 0.50 and 0.51, respectively, in the test cohort. In contrast, Path-IO demonstrated stronger discriminative ability, achieving C-indices of 0.69 for OS and 0.65 for PFS in the discovery cohort and 0.63 for OS and 0.58 for PFS in the test cohort.”

Furthermore, combining pathology-based predictions with radiomics and clinical data further improved the model’s ability to distinguish patient outcomes, with the C-index increasing from 0.58 to 0.70 for PFS and from 0.63 to 0.75 for OS. “These observations highlight the value of integrating multiple sources of information to guide treatment decisions,” said Bandyopadhyay.

Bandyopadhyay explained that, unlike prior pathomics studies, Path-IO is a biology-guided approach that grounds the predictions in tissue structures that are familiar to clinicians to reflect how pathologists naturally interpret tissue.

The model’s predictions correlated with immune profiling and multiplex imaging data because higher risk scores assigned by the model corresponded to phenotypes that are less likely to be sensitive to immunotherapy. “This correlation provides biological evidence for why certain patients may have better or worse outcomes with immunotherapy,” added Bandyopadhyay.

Bandyopadhyay emphasized that, since this approach was designed to be applied to routine pathology slides, if validated as a predictive tool, it could be incorporated into existing clinical workflows without significant expense compared to other emerging data-based technologies.

“This study represents, to our knowledge, the first deep learning-based pathomics biomarker rigorously validated across international real-world cohorts and a phase III randomized clinical trial, directly addressing one of the most urgent unmet needs in precision oncology: reliable patient selection and stratification for immunotherapy,” said Bandyopadhyay.

Limitations of the study, according to Bandyopadhyay, include that the study design is retrospective in nature and, although the results suggest that Path-IO may have predictive value in certain patient subgroups, further investigation is critically needed to go beyond the identification of patients who would benefit from immunotherapy and help predict what type of immunotherapy they can benefit from. “Future directions include prospective validation and the integration of paired, more comprehensive molecular profiling to enhance predictive performance and provide deeper molecular insights,” said Bandyopadhyay.

The study was supported by the National Institutes of Health, The UT MD Anderson Lung Moon Shot Program, and philanthropic contributions from Mrs. Andrea Mugnaini and Dr. Edward L.C. Smith, the Rexanna’s Foundation for Fighting Lung Cancer, QIAC Partnership in Research (QPR) funding, and Permanent Health Funds. Scientific and financial support for the Cancer Immune Monitoring and Analysis Centers and Cancer Immunologic Data Center (CIMAC-CIDC) Network was provided by the National Cancer Institute.

Bandyopadhyay declared no conflicts of interest.

Download a photo of Bandyopadhyay