Diagnostic Performance of Deep Learning Algorithms Applied to Three Common Diagnoses in Dermatopathology

Area of Science:

• Dermatopathology
• Artificial Intelligence
• Machine Learning

Background:

• Artificial intelligence (AI) is rapidly integrating into clinical settings, offering enhanced efficiency and accuracy in computer-aided diagnostics.
• Dermatopathology, reliant on pattern recognition, is well-suited for evaluating deep learning (DL) algorithms.

Purpose of the Study:

• To assess the diagnostic accuracy of DL algorithms for three prevalent dermatopathology conditions.
• To evaluate the performance of AI in classifying nodular basal cell carcinomas (BCCs), dermal nevi, and seborrheic keratoses.

Main Methods:

• Whole slide images (WSIs) of BCCs, dermal nevi, and seborrheic keratoses were annotated.
• A proprietary fully convolutional neural network was developed to train AI algorithms.
• WSIs with five distractor diagnoses were included in training sets to test algorithm robustness.

Main Results:

• The AI system achieved 99.45% accuracy for nodular basal cell carcinomas.
• Diagnostic accuracy for dermal nevi was 99.4%, and 100% for seborrheic keratoses.
• The AI demonstrated high precision in classifying the target dermatopathology diagnoses.

Conclusions:

• AI utilizing DL algorithms shows potential as a diagnostic aid in dermatopathology.
• This technology may lead to improved workflow efficiencies for dermatopathologists and laboratories.
• AI offers a promising adjunct for accurate and efficient dermatological diagnosis.