A hybrid framework for colorectal cancer detection and U-Net segmentation using polynetDWTCADx

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Oncology

Background:

• Colorectal cancer (CRC) diagnosis relies on accurate identification and segmentation of cancerous lesions.
• Existing methods may face challenges in achieving high precision for both classification and segmentation tasks.
• Advanced computational models are needed to improve diagnostic accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate a hybrid model, PolynetDWTCADx, for accurate identification and classification of colorectal cancer.
• To assess the efficacy of integrating Convolutional Neural Networks (CNNs), Discrete Wavelet Transforms (DWTs), and Support Vector Machines (SVMs) for enhanced feature extraction and classification.
• To evaluate the performance of U-Net architecture for semantic segmentation of cancerous colorectal regions.

Main Methods:

• A hybrid model, PolynetDWTCADx, was developed by combining CNNs, DWTs, and SVMs.
• DWT was employed to optimize and enhance two integrated CNN models.
• Classification was performed using SVM, and semantic segmentation utilized the U-Net architecture.

Main Results:

• PolynetDWTCADx achieved a testing accuracy of 92.3% and a training accuracy of 95.0%.
• The model demonstrated moderate recall and a high Area Under the Curve (AUC).
• U-Net achieved a maximal Intersection over Union (IoU) score of 0.93 for segmenting malignant colorectal tissues.

Conclusions:

• The PolynetDWTCADx model effectively distinguishes between noncancerous and cancerous colon lesions.
• The integration of DWT, CNNs, SVM, and U-Net provides detailed visual information crucial for diagnosis and treatment planning.
• PolynetDWTCADx shows significant potential to enhance the recognition and management of colorectal cancer.