DASNet: A Convolutional Neural Network with SE Attention Mechanism for ccRCC Tumor Grading.

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Clear cell renal cell carcinoma (ccRCC) is the most prevalent kidney cancer subtype.
• Advanced stages (III+) of ccRCC are associated with significantly higher mortality.
• Early detection is crucial for effective therapeutic intervention.

Purpose of the Study:

• To develop a non-invasive and efficient method for grading ccRCC using Computed Tomography (CT) images.
• To leverage deep learning and machine learning for accurate ccRCC classification.
• To enhance the early identification of ccRCC stages.

Main Methods:

• Utilized a Domain Adaptive Squeeze-and-Excitation Network (DASNet) for ccRCC classification.
• Employed MedAugment for dataset enhancement and balancing.
• Incorporated renal angiomyolipoma (AML) samples to prevent overfitting.
• Leveraged EfficientNet and RegNet as base models with Squeeze-and-Excitation (SE) attention.
• Applied Domain-Adversarial Neural Networks (DANNs) for domain consistency.

Main Results:

• Achieved a classification accuracy of 97.50% for ccRCC grading.
• Demonstrated the model's efficacy in identifying ccRCC grades from CT images.
• Showcased improved generalization and robustness through data augmentation and domain adaptation.

Conclusions:

• The proposed DASNet model offers a highly accurate, non-invasive approach for ccRCC grading.
• Findings support the clinical utility of deep learning in early kidney cancer detection.
• Establishes a foundation for wider application of AI in oncological imaging analysis.