Adipose Tissue Segmentation in Unlabeled Abdomen MRI using Cross Modality Domain Adaptation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Abdominal fat quantification is crucial for assessing health risks.
• Computed tomography (CT) is effective but uses ionizing radiation.
• Magnetic resonance imaging (MRI) offers superior soft tissue contrast but requires labor-intensive segmentation.

Purpose of the Study:

• To develop an automated algorithm for quantifying abdominal fat from MRI scans.
• To overcome the labor-intensive nature of manual fat segmentation in MRI.
• To enable accurate fat quantification using MRI without manual labeling.

Main Methods:

• A deep learning algorithm employing a cycle generative adversarial network (C-GAN) was developed.
• The algorithm performs cross-modality adaptation, transforming MRI scans into synthetic CT (s-CT) images.
• This transformation facilitates fat segmentation using Hounsfield unit (HU) values.

Main Results:

• The developed algorithm automates fat quantification from MRI scans.
• The method achieved an average success score of 3.80/5 for visceral fat and 4.54/5 for subcutaneous fat.
• Qualitative evaluation by expert radiologists confirmed the effectiveness of the segmentation.

Conclusions:

• The proposed deep learning approach offers an automated and efficient method for abdominal fat quantification using MRI.
• Cross-modality adaptation with C-GANs provides a viable solution for MRI fat segmentation without manual labeling.
• This technique holds promise for improved clinical assessment of abdominal adiposity.