.png)
There is a need to improve risk stratification of primary cutaneous melanomas to better inform and guide adjuvant treatment. Taking into account that hematoxylin and eosin (HE)-stained tumor tissue contains a large amount of clinically unexploited relevant morphological information, we evaluate AI-based tool, SmartProg-MEL, to predict survival outcomes in stages early melanoma patients from HE-stained whole-slide image (WSI).
SmartProg was evaluated on a cohort of 50 primary cutaneous melanomas (IHP-MEL-3, 72% IA, 12% IB and 16% IIA) with complete follow-up over 5 years. The model assigns a high or low risk profiles to patients only based on the primary tumor WSI. The stratification is assessed based on the model capability to predict relapses and OS among high profile risks as well as the Kaplan-Meier curves for the two profiles.
SmartProg-MEL predicts 38 patients as low risks (79% IA, 13% IB and 8% IIA) and 12 as high risks (50% IA, 8% IB and 42% IIA). Among the 7 deceased and 4 relapsed patients during the follow-up period, SmartProg assigned 4 (57%) and 2 (50%), respectively, to high risk profiles, see Fig. 1. Moreover, when looking at the survival curves, we can see that the two profiles are clearly separable and give a 5-year overall survival of 62% and 92%, respectively, for high and low risks (Fig. 2). Finally, we explore the composition of two risk profiles in terms of their clinicopathological factors.
The AI-based stratification algorithm, SmartProg-MEL, demonstrates promising performance in identifying distinct risk profiles in early-stage melanoma patients based solely on HE-stained WSI. The ability to differentiate between high and low-risk groups with clearly separable survival curves indicates that this tool could play a valuable role in improving prognosis and guiding therapeutic decisions.
Jérôme Chetritt, Céline Bossard, Sanae Salhi, Yahia Salhi