Related Experiment Video
Updated: Jul 1, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Pan-cancer image segmentation based on feature pyramids and Mask R-CNN framework
Juan Wang1, Jian Zhou2, Man Wang1
1School of Computer and Information Engineering, Institute for Artificial Intelligence, Shanghai Polytechnic University, Shanghai, China.
Medical Physics
|March 4, 2024
Summary
This study enhances deep learning cancer image segmentation using feature pyramids, improving accuracy by 4% for better multi-scale object detection in pathology images.
Area of Science:
- Medical Image Analysis
- Artificial Intelligence in Oncology
- Computational Pathology
Background:
- Cancer poses significant health and economic burdens, necessitating advanced diagnostic tools.
- Deep learning-based image segmentation is crucial for cancer detection and diagnosis.
- Existing segmentation frameworks struggle with multi-scale object segmentation efficiency.
Purpose of the Study:
- To improve deep learning segmentation framework performance for cancer images.
- To enhance the average precision (AP) index in segmentation tasks.
- To achieve effective multi-scale cooperation in target segmentation.
Main Methods:
- Utilized the Pan-Cancer Histology Dataset for Nuclei Instance Segmentation and Classification (PanNuke).
- Employed the Mask Region Convolutional Neural Network (Mask R-CNN) framework with an improved loss function.
- Integrated a feature pyramid processing scheme for enhanced feature extraction and multi-scale analysis.
Main Results:
- Achieved an average precision (AP) of 0.269 on the PanNuke dataset.
- Demonstrated a performance improvement of approximately 4% compared to the standard Mask R-CNN framework.
- Successfully addressed limitations in segmenting objects of varying sizes.
Conclusions:
- Feature pyramid processing is an effective strategy for improving medical image segmentation.
- The proposed method offers a feasible approach to enhance cancer image analysis.
- The findings support the advancement of AI-driven tools for pathological diagnosis.
