A NEW machine learning-based tool developed by researchers could revolutionise the way doctors diagnose tumours of unknown origin, which are notoriously difficult to treat and often result in poor patient survival. The team created a random forest machine learning classifier designed to identify the primary site of origin for tumour patients. This classifier was trained using a vast database of both primary and metastatic tumour samples.
In trials using both publicly available and internal validation samples, the classifier demonstrated remarkable accuracy, correctly identifying the site of origin in 97% of cases. Additionally, 85% of samples received high probability scores of 0.9 or greater, further validating the effectiveness of the technology.
One of the key innovations of this classifier is its ability to categorise tumours into 46 distinct origin sites or disease classes. This classification process, which combines the expertise of pathologists with t-SNE visualisation, also uncovered several new and uncharacterised disease classes, suggesting potential avenues for future research.
This breakthrough represents a major advancement in the clinical management of tumours of unknown origin, offering a more accurate, data-driven approach to diagnosing and treating patients with these complex tumours. By improving the ability to pinpoint the origin of a tumour, this machine learning classifier could significantly enhance treatment decisions and ultimately improve patient outcomes.
Helena Bradbury, EMJ
Reference
Duckett D et al. Accurate identification of primary site in tumors of unknown origin (TUO) using DNA methylation. Npj Precision Oncology. 2025;9(1):8.
