Predicting Response to Immune Checkpoint Inhibitors in Melanoma: Emerging Approaches in Digital Pathology, Spatial Profiling and Machine Learning

Jakub Banaszek, Dawid Bąk, Kinga Barańska, Alicja Czajka, Dominika Ciesielska, Jakub Kleinrok, Weronika Pająk, Agnieszka Korolczuk, Maciej Mazur, Kamil Rusztyn

The introduction of immune checkpoint inhibitors (ICIs) into the treatment of melanoma has significantly reduced mortality over the past decade. However, therapeutic benefit is not observed in all patients, and treatment may be associated with severe adverse events. Therefore, identifying patients who are most likely to benefit from immunotherapy remains of critical importance. Currently used biomarkers, such as programmed death-ligand 1 (PD-L1) expression and manual assessment of tumour-infiltrating lymphocytes (TILs), have limited predictive value. This narrative review provides a critical appraisal of studies employing digital pathology tools, multiplex and spatial techniques (including multiplex immunofluorescence, imaging mass cytometry, and digital spatial profiling), as well as machine learning algorithms for predicting response to ICIs in patients with melanoma. Available evidence suggests that the highest predictive value may be achieved by approaches integrating quantitative assessment of immune infiltration with information on its spatial distribution, functional state, and interactions within the tumour microenvironment. Particular relevance may be attributed to features associated with the “immune-inflamed”, “immune-excluded”, and “immune-desert” phenotypes, the presence of tertiary lymphoid structures, and the organisation of local immune niches. In addition, this review highlights key limitations in the interpretation of current data, including lack of methodological standardisation, data heterogeneity, and insufficient validation. Directions for future research necessary for the implementation of these approaches into routine clinical practice are also discussed.