Related Experiment Video
Updated: Sep 11, 2025
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
14.3K
Advancing deep learning-based segmentation for multiple lung cancer lesions in real-world multicenter CT scans
Xavier Rafael-Palou1, Ana Jimenez-Pastor2, Luis Martí-Bonmatí3,4
1Research and Frontiers AI, Quibim, Valencia, Spain. xavierrafael@quibim.com.
European Radiology Experimental
|August 18, 2025
Summary
This study introduces an automated method for segmenting multiple lung cancer lesions in CT scans, achieving 85% sensitivity in external validation. The AI-driven approach enhances lung cancer assessment and disease monitoring.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Oncology
- Radiomics and Computational Pathology
Background:
- Accurate segmentation of lung cancer lesions in computed tomography (CT) is critical for diagnosis, treatment planning, and response assessment.
- While single lesion segmentation is established, multi-lesion segmentation in lung cancer remains an underexplored area.
- This study addresses the need for automated multi-instance segmentation of lung cancer lesions using real-world multicenter data.
Purpose of the Study:
- To develop and validate a novel automated approach for multi-instance segmentation of lung cancer lesions in CT scans.
- To address the gap in automated segmentation of multiple lung cancer lesions per patient.
- To leverage a heterogeneous cohort with real-world multicenter data for robust model training and evaluation.
Main Methods:
- Analysis of 1,081 CT scans with 5,322 annotated lesions, stratified into training (n=868) and testing (n=213) subsets.
- Development of a three-step automated pipeline: thoracic bounding box extraction, multi-instance lesion segmentation, and false positive reduction using a multiscale cascade classifier.
- Utilized a deep learning-based segmentation pipeline trained on multi-center real-world data.
Main Results:
- Achieved a Dice similarity coefficient of 76% for segmentation and 85% lesion detection sensitivity on an independent test set.
- Validated on an external dataset of 188 cases, achieving a 73% Dice similarity coefficient and 85% lesion detection sensitivity.
- Demonstrated robust performance in detecting and segmenting multiple lung cancer lesions across different datasets.
Conclusions:
- The developed automated approach accurately detects and segments multiple lung cancer lesions per patient on CT scans.
- The method shows robustness and generalizability across independent and external real-world datasets.
- AI-driven segmentation of multiple lesions comprehensively captures lesion burden, improving lung cancer assessment and disease monitoring.
Keywords:
Artificial intelligenceDeep learningLung neoplasmsNeural networks (computer)Tomography (x-ray computed)
