Related Experiment Video
Updated: Jun 1, 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
14.1K
Automated Deep Learning-Based Detection and Segmentation of Lung Tumors at CT.
Mehr Kashyap1, Xi Wang2,3,4, Neil Panjwani2,5
1Stanford University School of Medicine, Department of Medicine, Stanford, CA, US.
Radiology
|January 21, 2025
Summary
An AI model accurately detects and segments lung tumors on CT scans, significantly reducing segmentation time compared to physicians. This deep learning approach aids in cancer monitoring and treatment planning.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Accurate lung tumor detection and segmentation on CT scans are vital for cancer management.
- Manual delineation is time-consuming and prone to inter-physician variability.
Purpose of the Study:
- To develop and assess an ensemble deep learning model for automated lung tumor identification and segmentation on CT scans.
Main Methods:
- A 3D U-Net-based, image-multiresolution ensemble model was trained on 1,504 CT scans with radiotherapy segmentation data.
- Performance was evaluated on internal and external test sets using sensitivity, specificity, and Dice Similarity Coefficient (DSC).
Main Results:
- The model achieved 92% sensitivity and 82% specificity in tumor detection.
- A median DSC of 0.77 was observed between model and physician segmentations, comparable to inter-physician variability (0.80).
- Automated segmentation was significantly faster (mean 76.6s) than manual delineation (166.1-187.7s).
Conclusions:
- Routinely collected radiotherapy data effectively trained the deep learning model.
- The 3D U-Net ensemble model demonstrates robust performance and generalizability for lung tumor segmentation.
