Extent of benefit from ensemble learning and patient metadata in dermatoscopic CNNs

Quantify the extent to which ensemble learning (i.e., aggregating predictions from multiple convolutional neural networks) and the incorporation of patient-specific metadata such as age and anatomical lesion location contribute to improving predictive performance in dermatoscopic skin disease classification.

Background

Prior work has explored improving dermatoscopic skin lesion classification by ensembling multiple CNNs and by adding patient-specific information (e.g., age, lesion site). Although these strategies are commonly reported, the magnitude of their contribution to overall model performance is not well established.

This study also investigates meta-classification and patient data integration but emphasizes that the precise benefit of these approaches remains unresolved, motivating a clear, quantitative assessment.