Classification of the Confocal Microscopy Images of Colorectal Tumor and Inflammatory Colitis Mucosa Tissue Using...

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Confocal microscopy image analysis aids neoplasm diagnosis but faces challenges with ambiguous cases.
• High inter-observer variability and extensive training times hinder manual analysis.
• Standardized, objective diagnostic tools are needed for accurate classification.

Purpose of the Study:

• To develop a deep learning model for classifying confocal microscopy images of colon tissues.
• The model aims to differentiate between neoplasm, inflammation, and normal tissue.
• To improve diagnostic accuracy and efficiency in pathological analysis.

Main Methods:

• Utilized ResNet50 architecture for image classification.
• Implemented data augmentation and transfer learning to train the model with limited data.
• Employed class activation mapping (CAM) for result interpretability.

Main Results:

• Achieved 81% accuracy in classifying colon tissue images.
• Outperformed traditional machine learning methods by 14.05%.
• Demonstrated superior performance compared to human endoscopists, exceeding their predictions by 22.6%.

Conclusions:

• Deep learning, specifically ResNet50 with augmentation and transfer learning, effectively classifies confocal microscopy images.
• The model enhances diagnostic accuracy for neoplasm, inflammation, and normal colon tissues.
• This approach can significantly reduce inter-observer variability and training time in pathology.