Early Melanoma Diagnosis With Sequential Dermoscopic Images

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Early melanoma diagnosis relies on dermoscopic image evaluation.
• Current algorithms use single images, missing crucial temporal changes.
• Ignoring lesion evolution can lead to misdiagnosis in borderline cases.

Purpose of the Study:

• To develop an automated framework for early melanoma diagnosis using sequential dermoscopic images.
• To capture and analyze temporal, morphological changes in skin lesions.
• To improve diagnostic accuracy and enable earlier detection of melanoma.

Main Methods:

• A three-step framework involving image alignment and difference computation.
• Utilizing a spatio-temporal network to analyze changes in aligned lesion and difference images.
• Developing an early diagnosis module to calculate malignancy probability scores over time.

Main Results:

• The proposed model demonstrated superior performance compared to existing sequence models.
• Achieved higher diagnostic accuracy (63.69%) than dermatologists (54.33%).
• Enabled earlier melanoma diagnosis, with 60.7% correctly identified on first follow-up images versus 32.7% by clinicians.

Conclusions:

• The framework effectively identifies melanocytic lesions at high risk of malignant transformation.
• The model redefines possibilities for early melanoma detection by analyzing lesion dynamics.
• Automated analysis of sequential dermoscopic images offers significant advantages in melanoma diagnosis.