Automatic Segmentation of Metastatic Livers Using U-Net
Author Information
Author(s): Tiraboschi Camilla, Parenti Federica, Sangalli Fabio, Resovi Andrea, Belotti Dorina, Lanzarone Ettore
Primary Institution: Department of Management, Information and Production Engineering, University of Bergamo
Hypothesis
Can neural networks based on U-net architecture effectively segment healthy and metastatic liver areas in micro-CT images of mice with pancreatic ductal adenocarcinoma?
Conclusion
The developed neural networks successfully segmented liver areas with performance comparable to manual methods, while saving time and ensuring reproducibility.
Supporting Evidence
- The networks achieved accuracy rates between 88.6% and 92.6% for different segmentation tasks.
- Specificity rates were above 93.8%, indicating reliable performance in distinguishing between healthy and metastatic tissues.
- The study represents a first step towards automated tools for liver metastasis research in preclinical settings.
Takeaway
Researchers created computer programs to help identify healthy and cancerous parts of the liver in mice, making it faster and easier than doing it by hand.
Methodology
Three U-net-based neural networks were developed and validated for segmenting healthy liver, metastatic liver area, and liver metastases in micro-CT images.
Limitations
The study was limited by the small number of animals used and focused only on advanced metastatic burden cases.
Participant Demographics
10-week-old female C57BL/6 mice were used in the study.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website