Scaling Immune Cell Quantification in Melanoma Through AI-Driven Assessment

According to Thazin Nwe Aung, PhD, an artificial intelligence (AI)–driven assessment for scoring tumor-infiltrating lymphocytes (TILs) may help better predict melanoma prognoses vs pathologist scoring. Aung, an associate research scientist in Pathology at the Yale School of Medicine, spoke with CancerNetwork® about the publication of a multi-institutional prognostic study she authored in JAMA Network Open that compared pathologist-read vs AI-driven assessments of TILs among patients with melanoma.

She began by highlighting the rationale of the study, which she explained was conducted to remedy potential inconsistencies that emerged during pathologist reads of TILs for melanoma. Despite pathologist scoring retaining value, a machine learning method for automated quantification of TILs was developed to overcome reader subjectivity and facilitate scalable and reproducible measurements.

Furthermore, she outlined the key findings from the study, which suggested that the AI-based assessment is more predictive of diagnosing melanoma. Aung concluded by further highlighting the scalability of the approach to quantify immune cells, which she expressed will help the risk stratification of the disease without major disruptions to routine workflows.

Data from the study revealed that the AI-based algorithm displayed superior reproducibility, with intraclass correlation coefficient (ICC) values higher than 0.90 for all machine learning TIL variables. Additionally, the AI-based scores showed prognostic associations with outcomes, with an HR of 0.45 (95% CI, 0.26-0.80; P = .005).

Transcript:

The rationale [of the study] was that although pathologist scoring of tumor-infiltrating lymphocytes is valuable, it is often subjective and inconsistent across [pathologists] and institutions because they look at the slides and give their best estimates. To address that variability, we developed a machine learning method that automatically counts or quantifies TILs and provides reproducible measurements at a multi-institutional scale.

The key finding from our study was the AI method is more reproducible than pathologist scoring, and [it may] better predict melanoma prognoses. Clinically, it offers a scalable way to quantify immune cells, which helps disease risk stratification and trial design without having to change these routine workflows.

Reference

Aung TN, Liu M, Su D, et al. Pathologist-read vs AI-driven assessment of tumor-infiltrating lymphocytes in melanoma. JAMA Netw Open. 2025;8(7):e2518906. doi:10.1001/jamanetworkopen.2025.18906