ViT-DCNN: Vision Transformer with Deformable CNN Model for Lung and Colon Cancer Detection.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Oncology
• Deep Learning for Cancer Diagnosis

Background:

• Lung and colon cancers are leading causes of mortality worldwide, with early detection posing a significant challenge.
• Histopathological images are crucial for accurate cancer diagnosis.
• Existing diagnostic methods require enhancement for improved accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for improved detection and classification of lung and colon cancers.
• To leverage the strengths of Vision Transformers and Deformable CNNs for enhanced feature extraction in medical images.
• To compare the performance of the proposed model against existing state-of-the-art models.

Main Methods:

• Utilized the Lung and Colon Cancer Histopathological Images Dataset, comprising five classes.
• Developed the Vision Transformer with Deformable CNN (ViT-DCNN) model, integrating self-attention with deformable convolutions.
• Trained and validated the model on training (80%), validation (10%), and test (10%) subsets.

Main Results:

• The ViT-DCNN model achieved high performance on the test set: 94.24% accuracy, 94.23% F1 score, 94.24% recall, and 94.37% precision.
• Demonstrated superior performance compared to ResNet-152, EfficientNet-B7, SwinTransformer, DenseNet-201, ConvNext, TransUNet, CNN-LSTM, MobileNetV3, and NASNet-A.
• Confirmed the model's robustness in detecting cancerous tissues through comprehensive evaluation.

Conclusions:

• The ViT-DCNN model offers a promising AI-driven solution for enhancing lung and colon cancer detection efficiency and reducing diagnostic errors.
• This model serves as a valuable tool for radiologists and clinicians, potentially transforming cancer diagnosis.
• Future research will focus on expanding the dataset and improving model interpretability for clinical application.