.png)
To refine prognostication in primary cutaneous melanoma (CM) and optimize adjuvant treatment decisions, we evaluate SmartProg-MEL (SPM), an AI-based histology-driven algorithm developed by DiaDeep and applied to H&E-stained whole-slide images (WSI) at the time of diagnosis. SPM provides a risk stratification score, under 15 minutes, for overall survival (OS) and relapse-free (RFS) outcomes to support clinical decision-making.
SPM was evaluated on a retrospective cohort of 383 primary CM with 5-year follow-up (46% IA, 15% IB, 9% IIA, 7% IIB, 6% IIC, 13% III, 4% IV). The model stratifies patients into high- or low-risk groups based solely on the WSI of the primary tumor. Kaplan-Meier curves are used to compare RFS and OS between risk groups, with statistical significance assessed using the log-rank test. A multivariable Cox regression analysis was performed to evaluate the independent prognostic value of SPM after adjusting for pathological factors. The negative and positive predictive values (NPV and PPV) of the SMP are explored.
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 tPatients with a low risk score had a significantly higher 5-y OS and RFS than patients of the high risk group (93.1% vs 62.5%, p<0.001 and 92.8% vs 47.1%, p<0.001). In multivariable analysis, SPM risk score was the strongest predictor (OS: HR=3.95, p<0.005, RFS: HR=5.03, p<0.005). In the I-IIA group, 29% (n=78) were assigned to the high risk profile with a decrease of the OS and RFS compared with the low risk group (OS: 95.4% vs 86%, p<0.05; RFS: 94.3% vs 74%, p<0.01). SPM has a NPV of 96%, 100% and a PPV of 17% and 69% in stages I/IIA and IIB/C respectively.he 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.
The AI-based risk stratification algorithm, SPM, demonstrates greater performance than stage in identifying high and low risk profiles, especially in early-stage CM patients. This new prognostic tool opens avenues for a routine clinical application to precise therapeutic decisions in an adjuvant setting.
Kanaan, Christina, Céline Bossard, Severine Roy, Naima Benannoune, Naima Hamoudi, Ines Besraoui, Magali Lacroix-Triki, Jérôme Chetritt, and Caroline Robert.