Developing AI-Based Assessments for TIL Scoring in Melanoma and Beyond

An artificial intelligence (AI)–driven assessment of tumor-infiltrating lymphocytes (TILs) may show superiority to pathologist-read scoring among patients undergoing evaluation for melanoma, according to Thazin Nwe Aung, PhD.

In a conversation with CancerNetwork® at Yale Cancer Center, Aung discussed the evaluation of the AI-driven assessment, as well as strategies for overcoming resistance to standard therapies for melanoma, based on 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.

An associate research scientist in Pathology at the Yale School of Medicine, Aung initially discussed the rationale for conducting the study, which she claimed attempted to address inconsistencies in pathologist scoring of TILs. She further expressed that her team developed the test to automate TIL quantification and provide a reproducible algorithm that could be scaled for multi-institutional use. Additionally, she outlined findings from the study, which revealed that the AI-driven assessment was more predictive of melanoma diagnoses vs pathologist reads and that its use may help standardize workflows.

Next, she outlined strategies for overcoming resistance to standard therapies for melanoma and other cancers, highlighting multimodal biomarker use to help predict treatment response and redirect patients to alternative therapies earlier in their treatment.Finally, she explained that the data and the algorithm have been released to enable outside testing, modification, and application of this system.

CancerNetwork: Regarding your study published in JAMA Network Open, what was the rationale for developing this algorithm and evaluating it vs traditional pathologist reading?

Aung: The rationale [of the study] was that although pathologist scoring of TILs 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.

What were the key findings and the clinical implications of the study? What might AI-driven lymphocyte quantification offer compared with traditional methods based on the data?

The key finding from our study was that 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.

What might the findings of the study suggest about the role that AI-based tools may play in the future of cancer detection and management?

Our study provided the evidence that AI can complement, not replace, the clinicians. In the future, these methods could be used to standardize the labor-intensive pathology workflow and reduce the variability; that can help [with] better personalized treatment strategies.

What can be done to overcome resistance to immunotherapy or other drugs among patients with melanoma or other malignancies?

Resistance happens because of these tumor cells. Tumor cells adapt and evade treatment. The strategy here is multimodal biomarkers; biomarkers that could be developed using various platforms, including transcriptomics, proteomics, and digital pathology. These biomarkers can predict which patients are not likely to respond to immunotherapy or any other treatments, or [it can predict] resistance to these treatments. These patients could be redirected to alternative therapies earlier.

What biomarkers may hold promise in the management and research of melanoma? What other factors may influence the treatment decision-making process for this population?

The current melanoma biomarkers are BRAF, NRAS, PD-L1, and tumor mutational burden. These TILs combined with clinical variables are the ones that are shaping melanoma or cancer management. Based on the measurement, or the intensity of the expression of these markers, the treatment decisions are being made.

Is there anything else related to your research in melanoma or work at Yale Cancer Center that you wanted to highlight?

We released our data from this published paper, [as well as] the algorithm [from the AI tool] that we developed, to accelerate the extent of validation and clinical implications so others can test, modify, and apply [them] at their own settings.

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