Related Experiment Video
Updated: Jun 22, 2025
07:53
Author Spotlight: Advancing 3D Modeling for Enhanced Diagnosis and Treatment of Pulmonary Nodules in Early-Stage Lung Cancer
Published on: October 13, 2023
1.4K
[Pulmonary PET /CT image instance segmentation based on dense interactive feature fusion Mask RCNN].
Tao Zhou1,2, Yanan Zhao1,2, Huiling Lu3
1School of Computer Science and Engineering, North Minzu University, Yinchuan 750021, P. R. China.
Summary
This study introduces a novel Dense Interactive Feature Fusion Mask RCNN (DIF-Mask RCNN) model to improve lung cancer detection and segmentation in PET/CT images. The model enhances feature extraction, leading to more accurate identification of tumor lesions.
Area of Science:
- Medical imaging analysis
- Artificial intelligence in oncology
- Computer-aided diagnosis
Context:
- Positron emission tomography/computed tomography (PET/CT) lung images present challenges like limited feature information, complex shapes, and blurred boundaries.
- Existing models struggle with inadequate extraction of tumor lesion features due to these image complexities.
Purpose:
- To propose a Dense Interactive Feature Fusion Mask RCNN (DIF-Mask RCNN) model to address the limitations in PET/CT lung image analysis.
- To enhance the extraction and segmentation of tumor lesion features, particularly weak and detailed information.
Summary:
- The DIF-Mask RCNN model incorporates a cross-scale feature extraction network and a dense interactive feature enhancement network.
- It utilizes a feature pyramid network (FPN) with dense connections to fuse shallow and deep features, improving the perception of subtle lesion characteristics.
- Experimental results on a clinical PET/CT lung image dataset demonstrate significant improvements in detection and segmentation accuracy compared to standard Mask RCNN.
Impact:
- The DIF-Mask RCNN model effectively detects and segments tumor lesions in PET/CT lung images.
- Achieved APdet of 67.16% and APseg of 68.12%, outperforming Mask RCNN (ResNet50) by 7.11% and 5.14% respectively.
- Provides a valuable tool and evaluation basis for computer-aided diagnosis of lung cancer.
