Multi-Input Dual-Stream Capsule Network for Improved Lung and Colon Cancer Classification

Area of Science:

• Digital pathology
• Computational oncology
• Artificial intelligence in medicine

Background:

• Lung and colon cancers are leading causes of human mortality.
• Accurate histopathological diagnosis is crucial for effective cancer treatment.
• Digital histopathology images offer a valuable resource for automated analysis.

Purpose of the Study:

• To develop an enhanced computerized diagnosis system for lung and colon cancers.
• To utilize a multi-input capsule network for improved detection of squamous cell carcinomas and adenocarcinomas.
• To leverage digital histopathology images for precise cancer classification.

Main Methods:

• A multi-input capsule network with two convolutional layer blocks (CLB and SCLB) was designed.
• CLB processed raw images, while SCLB processed pre-processed images (color balancing, gamma correction, sharpening, multi-scale fusion).
• A dual-input strategy was employed to enhance feature learning from pre-processed and raw histopathology images.

Main Results:

• The proposed model demonstrated significant improvements in classification accuracy on the LC25000 dataset.
• The system achieved cutting-edge performance across all tested cancer classes.
• An overall accuracy of 99.58% was obtained for detecting lung and colon abnormalities.

Conclusions:

• The multi-input capsule network effectively enhances computerized diagnosis of lung and colon cancers.
• The pre-processing techniques and dual-input approach contribute to superior feature learning.
• This AI-driven system shows great promise for accurate and efficient histopathological diagnosis.