Deep learning model for grading carcinoma with Gini-based feature selection and linear production-inspired feature...

Area of Science:

• Oncology
• Medical Imaging
• Computational Pathology

Background:

• Accurate grading of renal cell carcinoma (RCC) and hepatic cell carcinoma (HCC) is crucial for effective cancer treatment strategies.
• Traditional deep learning models face challenges in accurately classifying complex histopathology patterns in RCC and HCC.
• Existing methods often lack the precision needed for reliable cancer grading.

Purpose of the Study:

• To develop an advanced feature selection and fusion framework for enhanced grading of liver and kidney cancer.
• To improve classification accuracy for renal cell carcinoma (RCC) and hepatic cell carcinoma (HCC) using deep learning.
• To integrate attention mechanisms and economic theory-inspired methods for robust cancer grading.

Main Methods:

• Incorporation of attention mechanisms into MobileNetV2, DenseNet121, and InceptionV3 Convolutional Neural Network (CNN) architectures.
• Implementation of a Gini-based feature selection method to identify discriminative features from histopathology images.
• Optimal fusion of extracted features using an economic theory-inspired linear production function for improved prediction.

Main Results:

• The proposed framework achieved high classification accuracies: 93.04% for renal cell carcinoma (RCC) and 98.24% for hepatic cell carcinoma (HCC).
• Demonstrated superior performance compared to existing state-of-the-art models in cancer grading.
• Validated the robustness and effectiveness of the novel feature selection and fusion approach.

Conclusions:

• The developed attention-based CNN framework with economic theory-inspired feature selection and fusion significantly enhances the accuracy of RCC and HCC grading.
• This approach offers a promising tool for improving diagnostic precision in liver and kidney cancer.
• The publicly available code facilitates further research and application in computational pathology.