| Title:
|
Automatic Segmentation of Multiple Sclerosis Lesions from Brain MRI Using a Multi-Orientation TransUNet Model
|
| Author(s):
|
1
|
| Subject:
|
Advanced Techniques in Signal Acquisition and Image Processing
|
|
|
|
Presenter Author:
|
|
| Name:
|
Zeinab Yazdani
|
|
|
|
Author:
|
|
| Name 1:
|
Zeinab Yazdani - Corresponding - Presenter
|
| Affiliation 1:
|
School of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, Iran
|
|
|
| Name 2:
|
Hamid Soltanian-Zadeh
|
| Affiliation 2:
|
CIPCE, School of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, Iran
|
|
|
| Name 3:
|
|
| Affiliation 3:
|
|
|
|
| Name 4:
|
|
| Affiliation 4:
|
|
|
|
| Name 5:
|
|
| Affiliation 5:
|
|
|
|
| Name 6:
|
|
| Affiliation 6:
|
|
|
|
| Name 7:
|
|
| Affiliation 7:
|
|
|
|
| Name 8:
|
|
| Affiliation 8:
|
|
|
|
| Name 9:
|
|
| Affiliation 9:
|
|
|
|
| Name 10:
|
|
| Affiliation 10:
|
|
|
|
Abstract
|
Original Research
|
|
introduction :
|
Multiple Sclerosis (MS) is an autoimmune disease of the central nervous system that leads to the degradation of the myelin sheath of neurons in the brain and spinal cord. These lesions appear as plaques in MRI scans and are observed as hyperintense regions in FLAIR sequences. Manual segmentation of MS lesions is a time-consuming process. Therefore, the use of image processing techniques and artificial intelligence can significantly improve the accuracy and quality of MS lesion detection and segmentation.
|
|
Methods:
|
In this study, the TransUNet model, which combines the strengths of convolutional neural networks and transformer-based architectures within a U-Net framework, is employed. The experiments are conducted on the MSSEG1 dataset, consisting of 15 training samples and 38 test samples, which have been preprocessed in advance. Three-dimensional FLAIR images are converted into two-dimensional slices along the axial, coronal, and sagittal orientations and used for model training. During the testing phase, the segmentation results from each orientation are fused using a majority voting strategy to generate the final segmentation output.
|
|
Results:
|
The performance of the TransUNet model was evaluated under several configurations. In the baseline setting, the model achieved a Dice score of 48.96%. By modifying the optimizer and loss function, the Dice score improved significantly to 57.86%. Further improvement was obtained by training the model on two-dimensional slices extracted from three orientations (axial, coronal, and sagittal), resulting in a Dice score of 62.33%. Finally, combining multi-orientation training with a majority voting scheme during testing yielded the best performance, achieving a Dice score of 70.00%. These results demonstrate the positive impact of each proposed modification on MS lesion segmentation accuracy.
|
|
Conclusion:
|
The results of this study indicate that appropriate hyperparameter selection and multi-orientation slice-based training enhance the learning capability of the model and improve MS lesion segmentation performance. Moreover, the use of majority voting during the testing phase increases the reliability of lesion detection. The proposed approach can therefore be considered an effective method for automated MS lesion segmentation in MRI images.
|
| Attachment:
|
|
| Keywords:
|
Multiple Sclerosis, MS, Deep Learning, MRI, Segmentation, FLAIR sequence
|
| Key message:
|
|
| Do you want your abstract to be indexed, if this is selected as one of the top 100
articles?
|
