Integrating pathomics and deep learning for subtyping uveal melanoma: identifying high-risk immune infiltration profiles.

Purpose: Uveal melanoma (UVM) is the most common primary intraocular malignancy in adults, characterized by high mortality despite its relatively low incidence. This study aimed to utilize unsupervised learning techniques to identify a high immune infiltration subtype of UVM and improve patient stratification based on mortality risk. Methods: A total of 70 hematoxylin and eosin (H&E) stained whole-slide images (WSIs) of UVM were collected from the Genomic Data Commons (GDC) data portal, along with genomic and clinical data. An additional validation cohort of 68 UVM patients from West China Hospital was included. Pathomic features were extracted using CellProfiler software, and deep learning models were constructed for classification and survival prediction. Unsupervised clustering was performed to identify critical regions for prognosis prediction and patient classification. The relationship between histopathological features and genomics was explored. Results: The study achieved accurate prediction and classification of UVM patients using deep learning models and machine learning techniques. A high immune infiltration subtype of UVM was identified, which showed prognostic relevance. Unsupervised clustering categorized UVM patients into three distinct subgroups. The developed deep learning model based on the Inception-V3 architecture demonstrated promising results in survival prediction. Conclusion: This study demonstrates the potential of unsupervised learning and deep learning techniques in identifying a high immune infiltration subtype of UVM and improving patient stratification based on mortality risk. This research contributes to the field of computational pathology and highlights the potential of utilizing histopathological images, genomic data, and deep learning models in enhancing the management of UVM patients.