HGSC is an aggressive form of ovarian cancer. Some HGSCs are characterized by a defect in the homologous recombination pathway of DNA repair. This defect makes them sensitive to treatment with a class of targeted drugs called PARP inhibitors. It is critical to identify which HGSCs have HR deficiency in order to identify which patients should be offered treatment with PARP inhibitors. Currently, there are molecular tests which are used to detect HR deficiency, however they are not widely accessible and are often expensive and time-consuming to perform. This can delay access to therapy for patients. It is known that HGSCs with HRD have slightly different microscopic features to those without HR deficiency. We hypothesize that these differences can be reliably identified by an artificial intelligence algorithm trained on histologic slides from HGSCs. Such an algorithm would allow for tumours to be quickly pre-screened for the likelihood of HRD, and those tumours which are likely to be HRD can be prioritized for confirmatory molecular testing.
PARP inhibitor therapy for HGSCs represents a breakthrough in targeted treatment of these aggressive tumours. They are particularly effective in tumours with HRD, and therefore timely and accurate HRD testing is critical to ensure patients are offered the correct therapy. Currently available molecular assays for HRD testing are highly accurate but are often time-consuming to run and few labs possess the technical capabilities to offer them. Often, clinicians must send tumour tissue to commercial laboratories for testing, which introduces delays in turnaround time and expense to patients (in the case of unfunded tests). By exploiting the morphologic differences between HR-deficient and HR-proficient HGSCs, an AI algorithm capable of classifying tumour HRD status through examination of histologic slides alone would represent a fast and inexpensive adjunct to traditional molecular methods. This would enable clinicians to prioritize cases likely to harbour HR-deficiency for confirmatory testing, and a subset of cases could potentially be ‘screened out’ from requiring any molecular testing, thereby reducing turnaround times as well as costs to the healthcare system and patients. Additionally, while the goal of this algorithm would not be to completely replace molecular-based HRD testing, it could serve as a primary testing method in low- and middle-income jurisdictions which otherwise have no access to advanced molecular laboratory infrastructure, opening new avenues for access to targeted cancer therapy in these countries.