A Custom Image Recognition Model in the Cloud for Skin Diseases Identification.

Area of Science:

• Dermatology and Artificial Intelligence
• Computational Pathology
• Medical Imaging Analysis

Background:

• Increasing prevalence of skin diseases like solar lentigo and actinic keratosis due to global warming and UV radiation exposure.
• Need for efficient and accurate diagnostic tools for common dermatological conditions.
• Potential of machine learning in medical image classification.

Purpose of the Study:

• To propose a technical approach for classifying solar lentigo and actinic keratosis using Azure Custom Vision services.
• To leverage cloud-based AI for dermatological condition identification.
• To develop a convolutional neural network (CNN) model for image-based diagnosis.

Main Methods:

• Utilized Azure Custom Vision services for developing a CNN model.
• Trained the model on a dataset of 600 images sourced from the ISIC database.
• Employed manual image labeling for dataset preparation.

Main Results:

• Achieved a classification performance of 94% for the trained neural network on pre-trained images.
• Demonstrated the effectiveness of Azure Custom Vision in generating a high-performing CNN model with a relatively small dataset.
• Successfully developed a functional model for classifying solar lentigo and actinic keratosis.

Conclusions:

• The proposed technical approach using Azure Custom Vision is effective for classifying solar lentigo and actinic keratosis.
• Cloud-based AI services offer significant computational power and efficiency for medical image analysis.
• Manual image labeling remains a limitation in developing AI diagnostic tools, but high performance is achievable.