Automated Lung and Colon Cancer Classification Using Histopathological Images

Area of Science:

• Medical image analysis
• Computational pathology
• Artificial intelligence in oncology

Background:

• Cancer remains a leading global cause of mortality, with lung and colon cancers being significant contributors to death and disease.
• Accurate and efficient classification of histopathological images is crucial for timely cancer diagnosis and treatment planning.

Purpose of the Study:

• To develop an automated system for classifying lung and colon cancers using histopathological images.
• To evaluate the performance of a Swin Transformer V2 model against other deep learning architectures for this classification task.

Main Methods:

• Utilized the LC25000 dataset comprising histopathological images of lung and colon cancers.
• Developed a deep neural network classification system centered around the Swin Transformer V2 model.
• Employed data splitting, on-the-fly image augmentation, training, validation, and 5-fold cross-validation for robust performance evaluation.
• Assessed model performance using metrics including Accuracy, Kappa, confusion matrix, precision, recall, and F1 score.

Main Results:

• The Swin Transformer V2 model achieved a perfect score (100% Accuracy) across all evaluated metrics.
• This represents the first single model to attain perfect classification results on the LC25000 dataset.
• Comparative experiments demonstrated the superior performance of the Swin Transformer V2 over mainstream convolutional neural networks and other vision transformers.

Conclusions:

• The developed automated classification system, powered by the Swin Transformer V2, shows exceptional performance in distinguishing lung and colon cancers from histopathology images.
• This AI-driven approach has the potential to serve as a valuable tool to assist pathologists, enhancing the accuracy and efficiency of cancer diagnosis.