Gastric Pathology Image Classification Using Stepwise Fine-Tuning for Deep Neural Networks.

Area of Science:

• Medical Image Analysis
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Deep learning, particularly Convolutional Neural Networks (CNNs), excels at image classification.
• CNNs are crucial for advanced Computer-Aided Diagnosis (CAD) in pathology.
• A significant challenge is the scarcity of well-annotated pathology data due to high annotation costs.

Purpose of the Study:

• To develop a novel deep learning scheme to enhance gastric pathology image classification.
• To address the issue of insufficient annotated data for training deep neural networks.
• To improve the performance of state-of-the-art deep neural networks in pathology.

Main Methods:

• A novel stepwise fine-tuning-based deep learning scheme was proposed.
• New target-correlative intermediate datasets were established.
• The scheme was tested on several state-of-the-art deep neural networks using both standard and intermediate datasets.

Main Results:

• The proposed scheme effectively boosts the performance of deep neural networks for gastric pathology classification.
• The method demonstrates superiority compared to traditional approaches when tested.
• Experiments confirmed the feasibility of the scheme in improving classification accuracy.

Conclusions:

• The novel deep learning scheme successfully enhances gastric pathology image classification.
• The approach mitigates the problem of limited annotated data by mimicking pathologist perception.
• This method offers a cost-effective way to acquire pathology-related knowledge and improve CAD systems.