Related Experiment Video
Updated: Dec 30, 2025
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
7.3K
Feedback-based Self-improving CNN Algorithm for Breast Cancer Lymph Node Metastasis Detection in Real Clinical
Summary
This study introduces a feedback loop for digital pathology, enhancing classification algorithms with pathologist input. This novel approach significantly improves diagnostic accuracy in computational pathology.
Area of Science:
- Computational pathology
- Medical artificial intelligence
- Image analysis
Background:
- Digital pathology integrates classification algorithms, Graphical User Interfaces (GUIs), and pathologists.
- Current systems feature unidirectional interaction from algorithms to pathologists.
- Improving algorithm performance necessitates incorporating pathologist expertise.
Purpose of the Study:
- To introduce a novel feedback-based method for digital pathology.
- To enhance the performance of classification algorithms using pathologist feedback.
- To develop a simple and adaptive GUI for this feedback system.
Main Methods:
- Implementation of a bidirectional interaction pathway between algorithms and pathologists.
- Development of an adaptive Graphical User Interface (GUI) for seamless feedback.
- Application of the feedback method to a Convolutional Neural Network (CNN) algorithm.
Main Results:
- Significant improvement in classification performance was observed.
- The 25% quantile of prediction probability scores increased from 0.48 to 0.89.
- The median prediction probability score rose from 0.95 to 0.99.
Conclusions:
- The proposed feedback-based method effectively enhances digital pathology algorithm performance.
- Pathologist feedback is crucial for refining computational pathology tools.
- The developed GUI facilitates efficient integration of human expertise into AI models.
