Improving CNNs classification with pathologist-based expertise: the renal cell carcinoma case study.

Area of Science:

• Oncology
• Pathology
• Artificial Intelligence

Background:

• Accurate renal cell carcinoma (RCC) subtyping is crucial for patient prognosis and treatment decisions.
• Current histological slide analysis for RCC subtyping is time-consuming, subjective, and requires expert pathologists.
• Automated classification methods are needed to improve the efficiency and accuracy of RCC subtyping.

Purpose of the Study:

• To investigate the efficacy of deep learning methodologies for automatic RCC subtyping.
• To develop and evaluate a novel hybrid classification model (ExpertDeepTree) integrating deep learning and pathologist expertise.
• To compare the performance of the hybrid model against traditional deep learning approaches for RCC subtyping.

Main Methods:

• Utilized deep learning, specifically Convolutional Neural Networks (CNNs), for automated classification of RCC subtypes.
• Developed ExpertDeepTree (ExpertDT), a hybrid model combining supervised CNNs with pathologist expertise.
• Classified 91 patient samples diagnosed with clear cell RCC, papillary RCC, chromophobe RCC, or renal oncocytoma.

Main Results:

• Standard CNNs demonstrated variable performance across different RCC subtypes.
• The proposed ExpertDT model achieved superior performance in RCC subtyping compared to traditional CNNs.
• ExpertDT's effectiveness highlights the value of integrating expert knowledge into deep learning models for complex classification tasks.

Conclusions:

• Deep learning holds promise for automated RCC subtyping, but performance can be improved.
• A hybrid approach, ExpertDT, effectively combines AI and human expertise for enhanced RCC classification.
• Incorporating expert knowledge into AI models offers a valuable strategy for improving diagnostic accuracy in complex pathological cases.