AI spots DNA repair flaw in ovarian tumor images, in lab study

In a laboratory at King's College London, a computer has been taught to read what a microscope shows but a pathologist may not easily see: the faint trace of a broken DNA repair system inside ovarian cancer cells. The model, called HRDPath, works from high-resolution images of tissue slides, not from DNA. In a study published in Nature Communications, the team reports that it correctly identified approximately 91% of HRD-positive cases.

The target is homologous recombination deficiency, or HRD, a defect that is one of the most important features of high-grade serous tubo-ovarian carcinoma, an aggressive cancer arising in the ovaries or fallopian tubes. Cancer cells with HRD, which can come from mutations in genes such as BRCA1 and BRCA2, repair damaged DNA poorly. That weakness makes them more vulnerable to platinum-based chemotherapy and to a class of drugs called PARP inhibitors, so knowing who carries it guides the choice of treatment. Today that knowledge comes from genetic analysis, which the researchers describe as expensive, time-consuming and hard to access in resource-limited health care settings.

HRDPath works in stages: it extracts small patches from each slide, turns each patch into a set of numerical features using a model called RetCCL, then learns to predict HRD status from those features. The researchers also noticed visible differences that the model appears to pick up: HRD-positive cancers tended to have rounder cell nuclei and more densely packed cancer cells in certain tumor regions. The authors suggest that defects in DNA repair leave physical traces in tumor tissue, traces that conventional examination may struggle to recognize.

The figure needs its context. The 91% comes from the study's own tests, and it is strongest when several images from the same patient are considered together. The researchers state that further research is needed to validate the model in larger patient populations, and they give no timetable for that work.

For patients, the potential gain is reach. If HRD status can be reliably read from tissue images that hospitals already produce, fewer women might need a separate genetic test before a treatment decision, and clinics with limited resources could still identify candidates for platinum-based chemotherapy and PARP inhibitors. That future depends on the larger validation the team has yet to carry out, and the study does not announce any clinical deployment.