Predicting cutaneous malignant melanoma patients' survival using deep learning: a retrospective cohort study

Area of Science:

• Oncology
• Bioinformatics
• Artificial Intelligence in Medicine

Background:

• Cutaneous malignant melanoma (CMM) presents a poor prognosis, particularly in metastatic stages.
• Accurate prognostic prediction is crucial for guiding clinical management of CMM.

Purpose of the Study:

• To develop and validate a deep learning survival model for predicting overall survival in CMM patients.
• To assess the model's performance across different patient cohorts.

Main Methods:

• Utilized the Surveillance, Epidemiology, and End Results database for CMM patient data.
• Developed a deep learning survival model (DeepCMM) trained on a 2010-2013 cohort.
• Validated the model internally on 2014 data and externally on 2015 data.

Main Results:

• DeepCMM achieved an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.8270 in the training cohort.
• The model demonstrated consistent performance with AUCs of 0.8274 in the validation cohort and 0.8303 in the test cohort.
• DeepCMM was successfully packaged into user-friendly software for clinical application.

Conclusions:

• The DeepCMM model provides reliable predictions for cutaneous malignant melanoma patient survival.
• This deep learning approach can significantly aid in the prognostic assessment of CMM.