Deep scattering convolution network based features for ultrasonic fatty liver tissue characterization.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computer Science

Background:

• Excess liver fat accumulation causes liver dysfunction, potentially leading to fibrosis and cirrhosis.
• Early detection of fatty liver disease is vital to prevent irreversible liver damage.
• Liver tissue fat concentration is graded as Normal, Grade 1, Grade 2, and Grade 3.

Purpose of the Study:

• To develop an automated algorithm for grading fatty liver disease.
• To utilize texture analysis of ultrasound images for fatty liver classification.
• To leverage invariant scattering convolution network (ISCN) for feature extraction.

Main Methods:

• An automated algorithm was developed for grading fatty liver tissue.
• Invariant Scattering Convolution Network (ISCN) was employed for feature extraction.
• Summed scattering coefficients (SC) and a cubic Support Vector Machine (SVM) classifier were used.

Main Results:

• The ISCN generated stable, invariant representations of liver tissue texture.
• Summed SC features provided a compact representation for classification.
• The algorithm achieved 96.6% accuracy in categorizing fatty liver content.

Conclusions:

• Automated grading of fatty liver disease is feasible using texture analysis.
• ISCN provides effective features for classifying liver fat content from ultrasound images.
• The proposed method demonstrates high accuracy for early fatty liver detection.