Semi-Supervised Nests of Melanocytes Segmentation Method Using Convolutional Autoencoders.

Area of Science:

• Dermatopathology
• Computer Vision
• Deep Learning

Background:

• Accurate segmentation of nevus cell nests is crucial for diagnosing skin lesions.
• Distinguishing benign from malignant lesions relies on border irregularity criteria.
• Existing methods struggle with limited ground-truth data for segmentation tasks.

Purpose of the Study:

• To develop a semi-supervised segmentation solution for histopathological images.
• To address the challenge of limited ground-truth data in segmentation tasks.
• To accurately segment nevus cell nests for improved dermatopathology diagnostics.

Main Methods:

• Utilized a convolutional autoencoder architecture for segmentation.
• Implemented a two-step learning process within the autoencoder framework.
• Applied the method to histopathological images of skin specimens for nevus cell nest detection.

Main Results:

• Achieved a Dice similarity coefficient of 0.81.
• Demonstrated a sensitivity of 0.76 and specificity of 0.94.
• The proposed deep learning tool effectively segments nevus areas, outperforming existing methods.

Conclusions:

• The developed semi-supervised segmentation approach is effective for histopathology.
• The method successfully segments nevus cell nests, even with small datasets.
• This computer-vision tool offers a novel solution for dermatopathology image analysis.