Adaptive Wavelet-VNet for Medical Image Segmentation
Author Information
Author(s): Qian Xiaoxue, Lu Weiguo, You Zhang
Primary Institution: University of Texas Southwestern Medical Center
Hypothesis
Can we improve segmentation accuracy in medical images using a single sample test time adaptation method?
Conclusion
The adaptive WaVNet significantly enhances image segmentation accuracy from unseen domains during test time through unsupervised learning and multi-objective optimization.
Supporting Evidence
- The adaptive WaVNet achieved a mean Dice coefficient of 78.10 ± 5.23% on the liver dataset.
- It outperformed state-of-the-art TTA methods in segmentation accuracy.
- The method showed significant improvements over baseline models without adaptation.
Takeaway
This study shows a new way to help computers better understand medical images by learning from just one picture at a time, making it easier to use in real hospitals.
Methodology
The study proposed a wavelet-VNet model that adapts to unseen target domains using a hybrid objective function during test time, optimizing model parameters based on unlabeled test samples.
Potential Biases
The model may overfit to the single sample used for adaptation, leading to potential inaccuracies.
Limitations
The method may struggle with larger domain shifts and requires further accuracy improvements for multi-organ segmentations.
Participant Demographics
The study utilized datasets from different clinical sites for liver and prostate segmentation.
Statistical Information
P-Value
p<0.001
Statistical Significance
p<0.001
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website