Vision Transformers-Based Deep Feature Generation Framework for Hydatid Cyst Classification in Computed Tomography...

Area of Science:

• Medical Imaging
• Parasitology
• Artificial Intelligence

Background:

• Hydatid cysts, caused by Echinococcus granulosus, present a significant public health concern due to severe complications and mortality.
• Accurate classification of hydatid cyst types is crucial for effective treatment and management.

Purpose of the Study:

• To introduce a novel deep feature generation framework (ViT-DFG) for enhanced hydatid cyst classification.
• To improve the accuracy of differentiating hydatid cyst types using vision transformer models.

Main Methods:

• The ViT-DFG framework involves image preprocessing, feature extraction via vision transformers, feature selection using iterative neighborhood component analysis, and classification.
• Performance was evaluated using k-nearest neighbor and multi-layer perceptron classifiers on a dataset of five cyst types, analyzed in three-class (active, transition, inactive) and five-class settings.
• 5-fold cross-validation and one-way ANOVA were employed for performance evaluation and statistical analysis.

Main Results:

• The ViT-DFG framework achieved high classification accuracies: 98.10% for three-class and 95.12% for five-class classifications.
• The proposed method demonstrated superior performance compared to existing techniques and individual vision transformer models.
• Statistical analysis confirmed significant improvements offered by the ViT-DFG framework (p < 0.05).

Conclusions:

• The ViT-DFG framework effectively combines vision transformers and feature selection for superior hydatid cyst classification.
• This approach holds significant potential for advancing medical image analysis and automated diagnostics in parasitology.
• The findings suggest improved clinical decision-making through more accurate and automated hydatid cyst identification.