.png)
Predictive and prognostic biomarkers represent the cornerstone of clinical oncology. Improving the prognostication of melanoma is needed to select patients for effective adjuvant therapies. Taking into account that tumor tissue contains a large amount of clinically relevant hidden information that is not fully exploited, we propose a weakly-supervised deep-learning approach on H&E-stained whole-slide image (WSI) to directly predict survival outcomes in melanoma patients.
We designed a deep neural (DL) network that extracts features from no-padding patches of WSI (tiles of size 512x512 pixels) using a self-supervised learning framework, which are then fed along with survival information into the DL network assigning a survival risk score to each WSI/patient. The model was trained and validated on 195 cases of primary melanoma diagnosed between 2015 and 2017 originating from the IHP Group and the French RicMel network. Performance was evaluated using cross-validations as well as an external set of 238 cases from the TCGA database. Concordance index (c-index) was used as a metric to assess the performance of the proposed approach.
For the prediction of survival outcomes, the proposed pipeline yielded a c-index of 0.75 and 0.66 for the IHP Group cohort and TCGA dataset respectively. Furthermore, we were able to significantly discriminate the 2 groups of patients, with a good and a poor prognosis in terms of survival with p<0.001 for IHP and p=0.01 for the TCGA based on Kaplan-Meier survival analysis. Our prediction model outperforms the reported scores for the other types of tumours based on deep learning frameworks.
This innovative weakly-supervised deep-learning approach to HE image analysis allows the deep-network to choose its own morphological features, and thus highlights new morphological prognostic biomarkers. Based on this prognostic risk score, we aim to accelerate and improve the clinical decision-making process in the field of melanoma. Due to this exiting result, the performance of our algorithm is challenge on a third independent dataset.
C Bossard, Y Salhi, G Quereux, A Khammari, S Salhi, C Jérôme