GASTRO-CADx: a three stages framework for diagnosing gastrointestinal diseases

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Gastrointestinal (GI) diseases pose significant diagnostic challenges, often involving expensive and complex procedures.
• Current diagnostic methods can be time-consuming and may lack the desired accuracy.
• Deep learning (DL) offers a promising avenue for developing efficient and accurate computer-aided diagnosis (CADx) systems.

Purpose of the Study:

• To propose and evaluate a novel DL-based CADx system, Gastro-CADx, for classifying various GI diseases.
• To investigate the effectiveness of combining different feature extraction techniques for improved diagnostic performance.
• To compare the proposed system's accuracy against existing methods in GI disease classification.

Main Methods:

• Gastro-CADx employs a three-stage approach: initial spatial feature extraction using Convolutional Neural Networks (CNNs).
• Subsequent extraction of temporal-frequency and spatial-frequency features using Discrete Wavelet Transform (DWT) and Discrete Cosine Transform (DCT), followed by feature reduction.
• Final stage involves fusing diverse feature sets to identify the optimal combination for classification.

Main Results:

• Gastro-CADx achieved high classification accuracies of 97.3% on Dataset I and 99.7% on Dataset II.
• The system demonstrated superior performance compared to recent related works in GI disease classification.
• Feature fusion strategies proved effective in enhancing the CADx system's diagnostic capabilities.

Conclusions:

• The proposed Gastro-CADx system offers a highly accurate and efficient method for classifying GI diseases using DL.
• This approach has the potential to significantly reduce healthcare costs, decrease mortality rates, and aid gastroenterologists in making faster, more precise diagnoses.
• Gastro-CADx represents a significant advancement in applying AI to gastroenterology diagnostics.