AI-driven eyelid tumor classification in ocular oncology using proteomic data

Abstract

Eyelid tumors pose diagnostic challenges due to their diverse pathological types and limited biopsy materials. This study aimed to develop an artificial intelligence (AI) diagnostic system for accurate classification of eyelid tumors. Utilizing mass spectrometry-based proteomics, we analyzed proteomic data from eight tissue types and identified eighteen novel biomarkers based on 233 formalin-fixed, paraffin-embedded (FFPE) samples from 150 patients. The 18-protein model, validated by an independent cohort (99 samples from 60 patients), exhibited high accuracy (84.8%), precision (86.2%), and recall (84.8%) in multi-class classification. The model demonstrated distinct clustering of different lesion types, as visualized through UMAP plots. Receiver operator characteristic (ROC) curve analysis revealed strong predictive ability with area under the curve (AUC) values ranging from 0.80 to 1.00. This AI-based diagnostic system holds promise for improving the efficiency and precision of eyelid tumor diagnosis, addressing the limitations of traditional pathological methods.