Designing an Intuitive Human-Machine Interface for a Skin Cancer Diagnostic System: An Ensemble Learning Approach

Abstract

advanced computational models for practical clinical application. This study introduces a refined Ensemble Learning-Based Decision Support System, designed with an emphasis on intuitive HMI for accurate melanocytic and non-melanocytic skin cancer diagnosis. We present ”EffiViT,” a model that synergizes EfficientNet’s robust feature extraction capabilities with the Vision Transformer’s attention-based contextual understanding, tailored through an interface that prioritizes ease of use and interpretability for medical professionals. Through extensive evaluation on the ISIC 2019 benchmark dataset, EffiViT demonstrated a classification accuracy of 99.4%, coupled with superior performance in specificity and area under the ROC curve. The system’s interface design was iteratively refined based on feedback from dermatologists, focusing on clear visualization of diagnostic information, straightforward navigation, and efficient access to model interpretations. Our findings underscore the importance of integrating user-centered design principles in the development of diagnostic tools, highlighting how a well-conceived HMI can enhance the adoption and effectiveness of AI-based systems in clinical settings. The proposed system stands out not only for its diagnostic accuracy but also for its contribution to the realm of HMI, offering insights into designing interfaces that facilitate better decision-making and ultimately improve patient outcomes in the field of dermatology.