Rethinking BCC diagnosis: Automating concept-specific detection of BCC in dermatoscopic images

Basal cell carcinoma (BCC) is a prevalent type of skin cancer in which the inherent subjectivity of dermoscopy poses diagnostic challenges. Existing AI systems, which provide mainly image-level insights, lack the interpretability that is crucial for effective clinical decisions and patient education.

Our study developed a refined BCC dataset from the Human‒Machine Adversarial Model (HAM10000), which was annotated by clinicians to identify key diagnostic features. We integrated the ResNet50 and Mask R-CNN architectures to enhance the model's performance by synthesizing feature-related knowledge. Statistical evaluations, such as grouped bar charts and line graphs, validated the improvement in our clinical diagnosis evaluation scheme.

The RFSD-BCC system significantly enhanced the diagnosis of BCC, with higher sensitivity, specificity, and accuracy. The system achieved an area under the precision-recall curve of 0.84, which closely matches physicians' diagnoses with high R² values and low MAEs. With the RFSD-BCC, the sensitivity increased by 7%, the specificity increased by 11%, the accuracy increased by 10%, and the intraclass correlation coefficient increased by 6%, which demonstrates the system's effectiveness in clinical settings.

The RFSD-BCC system improves BCC diagnosis by integrating feature combination models, which enhances both sensitivity and specificity. It offers interpretable diagnoses that bridge AI analysis with clinical practice, significantly improving clinicians' diagnostic accuracy and fostering better patient understanding.