NA-MIC Project Weeks
Back to Projects List
TorchXRayVision Meets 3D Slicer: Bridging Deep Learning and Medical Imaging
Key Investigators
- Constantin Constantinescu (Lucian Blaga University of Sibiu, Romania)
- Juan Ruiz-Alzola (University of Las Palmas de Gran Canaria, Spain)
- Csaba Pintér (EBATINCA, Spain)
- Oscar Martin (EBATINCA, Spain)
- Borja Fernandez (EBATINCA, Spain)
Project Description
This project focuses on developing a 3D Slicer module for the automatic processing of chest X-rays, integrating powerful deep learning capabilities provided by TorchXRayVision. The module streamlines radiological analysis by offering the following features:
- Segmentation: Automatically identify and outline anatomical regions in chest X-rays, such as lungs or other structures.
- Anomaly Detection: Detect abnormalities and highlight regions of interest for further investigation.
- Pathology Classification: Classify pathologies such as pneumonia, atelectasis, or other common conditions.
By combining the advanced machine learning models from TorchXRayVision with the versatile 3D Slicer platform, this module aims to provide a robust tool for clinicians and researchers to enhance diagnostic workflows, reduce manual workload, and improve consistency in radiological interpretation.
Objective
- A 3D Slicer Module
- TorchXRayVision models included in the module
- Torch XRays automatic segmentation, anomaly detection and pathology classification
- Heatmaps
Approach and Plan
- Create a Slicer Module
- Create an interface to upload X-Rays and perform automatic analysis
- Use TorchXRayVision framework to perform automatic analysis
- Compute heatmaps
Progress and Next Steps
- Creating a 3D Slicer Module
- Building the interface
- Including the TorchXRayVision models
- Incorporate mechanisms to facilitate the interpretability of the predictions made by the models
Illustrations
Loading the X-Ray Image:
Running automatic analysis:
View segmentation:
Pathologies scores and heatmap validation: