An AI model created by researchers at King's can detect signs of faulty repair mechanisms in ovarian cancer cells, potentially leading to a low-cost, widely accessible diagnostic tool in the future.
By analysing tissue images, the model could accurately recognise tumours that contained broken DNA repair mechanisms. This could enable clinicians to choose treatments that exploit this flaw.
We are unlocking information hidden in microscopic images of cancer tissue that could transform how we match patients to treatments. By using AI to reveal defects in DNA repair, we could make precision cancer medicine faster, more affordable and accessible to more women with ovarian cancer, not just in the UK but around the world."
Dr Heba Sailem, Reader in Computational Pathology at King's College London
The study focused on high-grade serous tubo-ovarian carcinoma (HGSOC), an aggressive form of cancer that arises in the ovaries or fallopian tubes. One of the most important characteristics of this cancer is something called homologous recombination deficiency, or HRD.
Cancer cells with HRD, which can arise from mutations in genes such as BRCA1 and BRCA2, have an impaired ability to repair DNA damage. This weakness can make cancer cells more vulnerable to certain treatments - in particular, platinum-based chemotherapy and a class of drugs called PARP inhibitors.
This makes HRD a valuable biomarker, as it helps doctors decide which treatments are likely to be effective. However, current HRD tests require sophisticated genetic analysis, which can be expensive, time-consuming and difficult to access in healthcare settings with fewer resources.
In the study, published in Nature Communications and led by Dr Heba Sailem, Reader in Computational Pathology at King's, the researchers developed a model called HRDPath. Rather than looking at DNA, HRDPath analysed high-resolution digital images of tissue samples under a microscope. The model was trained to recognise HRD-positive and negative cancers using tissue images from patients whose HRD status was already known through genomic analysis.
HRDPath demonstrated promising accuracy in identifying HRD-positive cancers, especially when considering several images from the same patient, correctly identifying approximately 91% of HRD-positive cases.
The researchers also identified subtle differences in the appearance and organisation of cancer cells associated with HRD. For example, HRD-positive cancers tended to have rounder cell nuclei and more densely packed cancer cells in certain tumour regions. The findings suggest that defects in DNA repair are associated with physical changes in tumour tissue that can be recognised by AI, and may be difficult to recognise through conventional examination.
The researchers believe HRDPath has the potential to make precision cancer medicine more accessible. If HRD status can be reliably identified using tissue images that hospitals already acquire for diagnosis, the technology could reduce reliance on expensive genetic testing and help clinicians identify patients who may benefit from treatments such as platinum-based chemotherapy and PARP inhibitors.
While further research is needed to validate the model in larger patient populations, HRDPath represents a promising step towards making personalised cancer treatment more widely available, helping more patients access therapies tailored to their individual tumours.
