.png)
In the field of ovarian cancer diagnosis, predicting Homologous Recombination Deficiency (HRD) holds paramount importance for personalized treatment strategies. Histopathology tissue analysis is considered as the gold standard in cancer diagnosis, prognosis, and theranostic. Whole Slide Imaging, i.e the scanning and digitization of entire histology slide, is now being adopted in numerous pathology labs, allowing development of deep-learning analysis of the large amount of morphological features contained in these WSI. The aim of this study is the development of a cutting-edge deep learning model designed to predict HRD status directly from WSIs of ovarian cancer tissue samples.
We introduce a DNN designed to predict the HRD status for patients with ovarian cancer from whole slide images (WSI), using a fusion-like model using both cellular information and tissue-level morphological features. The algorithm has been trained and evaluated using a cross-testing and cross-validation technique on 151 patients with at least one WSI from a discovery dataset using a 5-fold stratified framework. We then externally validated the model onto The Cancer Genome Atlas (TCGA) from which the HRD status is available (n=93).
The performance of HRD status prediction was assessed on the basis of area under curve (AUC).The proposed architecture achieved an AUC of 0.74 on the discovery cohort and 0.67 on the TCGA. Results to the pathologist could be provided immediately and integrated into the pathologist report. External validation on the PAOLA cohort dataset is ongoing.
By harnessing the power of deep neural networks (DNN), we provide a rapid and scalable solution for HRD prediction, circumventing the limitations of traditional molecular assays. Successful integration of this deep learning model into routine pathology workflows could significantly enhance diagnostic efficiency, reduce the turnaround time and financial cost compared with molecular assay. It could finally inform clinicians in tailoring targeted therapeutic interventions for ovarian cancer patients, thereby advancing precision medicine in the context of HRD-associated ovarian cancers.
Jean-Sebastien Frenel, Céline Bossard, Joseph Rynkiewicz, Florian Thomas, Yahia Salhi, Sanae Salhi, Jerome Chetritt