Related Experiment Video
Updated: Jun 13, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Accelerating segmentation of fossil CT scans through Deep Learning.
Espen M Knutsen1,2, Dmitry A Konovalov3
1College of Science and Engineering, James Cook University, Townsville, QLD, 4811, Australia. espen.knutsen@jcu.edu.au.
Scientific Reports
|September 9, 2024
Summary
Deep learning now automates fossil segmentation from CT scans with minimal data. This new method significantly reduces processing time for 3D fossil models.
Area of Science:
- Paleontology
- Computer Science
- Artificial Intelligence
Background:
- Automated segmentation of fossil CT scans is crucial for creating 3D models.
- Previous methods required extensive training data, limiting their application.
- Deep learning offers potential for efficient fossil data processing.
Purpose of the Study:
- To develop an automated Deep Learning segmentation method for fossil CT data.
- To train a model using a small fraction of the CT dataset (1-2%).
- To enable high-fidelity 3D model generation of fossils extracted from surrounding rock.
Main Methods:
- Implemented an automated Deep Learning segmentation workflow.
- Trained a Unet segmentation model with limited CT scan data.
- Validated the model's performance using Dice similarity.
Main Results:
- Achieved high-fidelity 3D models of fossil material.
- Successfully trained the Deep Learning model with less than 1-2% of the CT dataset.
- The final Unet model reached a validation Dice similarity score of 0.96.
Conclusions:
- This novel workflow significantly reduces processing time for CT-scanned fossil data.
- The method enhances the availability of segmented fossil material for research.
- It has the potential to revolutionize the use of Deep Learning in paleontology.
