Advancements in psoriasis classification using custom transfer learning algorithms.

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Psoriasis affects 2-3% of the global population, impacting quality of life and increasing risks of comorbidities.
• Current diagnostic methods for psoriasis subtypes lack consistent accuracy due to human error.
• Accurate classification of psoriasis is crucial for effective treatment and management.

Purpose of the Study:

• To develop an automated system for classifying seven types of psoriasis using deep learning.
• To create a novel, multi-class psoriasis dataset from publicly available sources.
• To address class imbalance issues in dermatological datasets through image augmentation.

Main Methods:

• Utilized publicly available datasets: SKIN LESION, ISIC, and DEMANET.
• Employed image augmentation techniques to balance the dataset classes.
• Implemented transfer learning with deep learning models: ResNet50, InceptionResNetV2, and InceptionV3.
• Trained and validated models using Adam and RMSprop optimizers.

Main Results:

• InceptionV3 achieved the highest accuracy (99.57% training, 96.82% validation, 98.68% testing).
• InceptionV2 also demonstrated strong performance (99.07% training, 96.65% validation, 97.20% testing).
• ResNet50 showed good, though comparatively lower, accuracy rates (92.36% training, 84.59% validation, 83.55% testing).

Conclusions:

• Deep learning models, particularly InceptionV3, show significant promise for accurate psoriasis classification.
• The developed dataset and methodology offer a robust tool for automated psoriasis diagnosis.
• This AI-driven approach can potentially improve diagnostic accuracy and overcome limitations of conventional methods.