Integrating Histopathology Foundation Models and PRAME Immunohistochemistry in Melanoma Diagnosis

Background: The histopathological diagnosis of challenging melanocytic lesions remains limited by substantial interobserver variability. Although computational pathology foundation models demonstrated promising diagnostic performance, their predictions are often difficult to interpret in terms of established criteria. Objective: We evaluated a diagnostic framework combining foundation-model-based whole-slide image analysis, automated prediction of morphological features, and manually assessed PRAME immunohistochemistry. Methods: Using the Virchow2 foundation model and STAMP framework, we analyzed a retrospective cohort of 1,000 melanocytic lesions undergoing PRAME immunohistochemistry. Models were trained on 800 cases and evaluated on a temporally held-out test cohort of 200 cases. Results: The slide-level malignancy classifier achieved a test AUROC of 0.970 (AUPRC 0.984). Separate models predicted established histopathological criteria with high discrimination, including ulceration (AUROC 0.991) and dermal mitoses (AUROC 0.966). In contrast, predicting PRAME expression from H&E images alone showed lower performance (AUROC 0.776-0.780 for PRAME scores 4 and 0). An integrated model combining the slide-level malignancy score, AI-predicted morphological features, and manual PRAME status achieved an AUROC of 0.972 (AUPRC 0.975), comparable to the standalone classifier. In multivariable analysis, PRAME score 4 versus 0 was strongly associated with malignancy (adjusted OR 35.04, 95% CI 17.39-74.62). Conclusions: These findings demonstrate the feasibility of combining foundation-model-based histopathological analysis with established morphological criteria and PRAME immunohistochemistry. Although the integrated framework did not substantially improve overall discrimination, it provides a structured approach for incorporating complementary histopathological and immunohistochemical information into AI-assisted assessment of melanocytic lesions.

Authors

Institutions

Publication Details

Journal
medRxiv
Published
2026-11-16
DOI
https://doi.org/10.64898/2026.11.14.26368089
Primary Topic
AI in cancer detection
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Integrating Histopathology Foundation Models and PRAME Immunohistochemistry in Melanoma Diagnosis

Jörg Laske, Mildred Sergon, Stefan Beissert, Claudia Günther et al.
medRxiv
AI in cancer detection
preprint

Integrating Histopathology Foundation Models and PRAME Immunohistochemistry in Melanoma Diagnosis

Jörg Laske, Mildred Sergon, Stefan Beissert, Claudia Günther, Sarah Hobelsberger, Daniela Ellen Aust, Narmin Ghaffari Laleh, Gustavo Bruno Baretton, Jakob Nikolas Kather, Frank Friedrich Gellrich, Cosima Hufnagel, Nick Reidow, Alaska Kistner
preprint en

Abstract

Background: The histopathological diagnosis of challenging melanocytic lesions remains limited by substantial interobserver variability. Although computational pathology foundation models demonstrated promising diagnostic performance, their predictions are often difficult to interpret in terms of established criteria. Objective: We evaluated a diagnostic framework combining foundation-model-based whole-slide image analysis, automated prediction of morphological features, and manually assessed PRAME immunohistochemistry. Methods: Using the Virchow2 foundation model and STAMP framework, we analyzed a retrospective cohort of 1,000 melanocytic lesions undergoing PRAME immunohistochemistry. Models were trained on 800 cases and evaluated on a temporally held-out test cohort of 200 cases. Results: The slide-level malignancy classifier achieved a test AUROC of 0.970 (AUPRC 0.984). Separate models predicted established histopathological criteria with high discrimination, including ulceration (AUROC 0.991) and dermal mitoses (AUROC 0.966). In contrast, predicting PRAME expression from H&E images alone showed lower performance (AUROC 0.776-0.780 for PRAME scores 4 and 0). An integrated model combining the slide-level malignancy score, AI-predicted morphological features, and manual PRAME status achieved an AUROC of 0.972 (AUPRC 0.975), comparable to the standalone classifier. In multivariable analysis, PRAME score 4 versus 0 was strongly associated with malignancy (adjusted OR 35.04, 95% CI 17.39-74.62). Conclusions: These findings demonstrate the feasibility of combining foundation-model-based histopathological analysis with established morphological criteria and PRAME immunohistochemistry. Although the integrated framework did not substantially improve overall discrimination, it provides a structured approach for incorporating complementary histopathological and immunohistochemical information into AI-assisted assessment of melanocytic lesions.

medRxiv
Universitätsklinikum Tübingen (DE), University Hospital Carl Gustav Carus (DE), Else Kröner Fresenius Center for Digital Health (DE), University of Tübingen (DE)
AI in cancer detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon