Hybrid Deep Learning Model for Endoscopic Lesion Detection and Classification Using Endoscopy Videos.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Gastroenterology

Background:

• Stomach disease detection is challenging due to symptom overlap and complex imaging.
• Computer-aided diagnosis (CAD) is crucial for efficient and accurate medical image analysis.
• Endoscopy videos present unique challenges for automated disease identification.

Purpose of the Study:

• To develop a hybrid method for accurate detection and classification of stomach diseases using endoscopy videos.
• To improve the diagnostic capabilities of computer-aided diagnosis systems in gastroenterology.
• To address the limitations of existing methods in analyzing complex endoscopic imagery.

Main Methods:

• A hybrid approach combining deep learning (VGG19, Alexnet) with transfer learning for feature extraction.
• Utilizing a genetic algorithm (GA) for adaptive feature selection and fusion.
• Employing multiple machine learning classifiers, including Cubic SVM, for final disease classification.
• Evaluation on a custom dataset encompassing gastritis, ulcer, esophagitis, bleeding, and healthy classes.

Main Results:

• The proposed hybrid method achieved a superb accuracy of 99.8% when using Cubic SVM.
• Comprehensive evaluation included classification accuracy, recall, precision, FNR, AUC, and processing time.
• The technique demonstrated superior performance compared to existing state-of-the-art methods.

Conclusions:

• The developed hybrid method offers a highly accurate and efficient solution for stomach disease detection and classification.
• This approach significantly enhances computer-aided diagnosis in medical imaging for gastroenterology.
• The fusion of deep learning features with genetic algorithm optimization shows great promise for clinical applications.