Advancing infantile hemangioma diagnosis by integrating temperature, color, and texture.

Abstract

Significance: Infantile hemangiomas are one of the most prevalent benign tumors in childhood. Typically, diagnosis relies on visual assessment of area, texture, and color. A few studies have focused on various color attributes in superficial and mixed Infantile hemangioma types, neglecting the deep category. Limited research has explored temperature in the location of hemangioma lesions.

Aim: We seek, for the first time, to quickly identify and classify infantile hemangioma lesions using a portable, programmable handheld device. The system aims to (1) replicate a physician's assessment of infantile hemangioma and (2) deliver an easy way to understand automatic diagnosis.

Approach: The custom-built device comprises an infrared sensor and a visible light spectrum sensor to assess color and depth through computations of different color and texture features. Over a 3-year period, 53 patients were monitored, and 83 hemangioma images were captured.

Results: The device accurately localized all lesions in real time and classified hemangioma lesions into three primary types using selected color and texture features. Evaluation metrics showed an average sensitivity of 0.8948 and specificity of 0.7313 for an accuracy of 0.7572 and an average sensitivity of 0.7803 and specificity of 0.8720 for an $F$ -score of 0.7826 in the three-class classification.

Conclusions: The two-sensor device accurately identifies and categorizes infantile hemangioma lesions, providing a clear automated diagnosis based on computerized features.