Beyond tumor tissue: Leveraging liquid biopsies and deep learning for prognostic insights in uveal melanoma

Uveal melanoma (UM) is a rare but aggressive eye tumor in adults. The course of the disease is often difficult to predict, and metastases are usually detected only at a late stage. This dissertation explores new methods to better monitor UM and stratify patient risk using liquid biopsies (blood tests that provide tumor information) and artificial intelligence.
Several studies demonstrated that tumor cells and tumor-derived DNA circulating in the blood are powerful predictors of disease progression. Circulating tumor DNA (ctDNA) was detected in the majority of patients with metastases and turned out to be the strongest predictor of overall survival. Genetic alterations of the primary tumor, such as chromosome 3 loss, could also reliably be identified in ctDNA. In addition, methylation analysis of circulating DNA revealed distinct risk groups and promising possibilities for future clinical application.
Beyond DNA, blood metabolites and inflammatory markers were analyzed. These provided additional insights into immune responses and tumor biology, clearly distinguishing patients from healthy controls. In parallel, deep learning models were developed that accurately predicted genetic mutations in UM, improving the classification of tumors based on imaging data.
Overall, the results show that liquid biopsy techniques and artificial intelligence represent valuable, non-invasive tools for diagnosis, prognosis, and treatment monitoring in UM. These innovations hold great promise for earlier detection of metastases, more personalized therapies, and ultimately, better care for patients affected by this challenging disease.