Classification for liver ultrasound tomography by posterior attenuation correction with a phantom study

Area of Science:

• Medical Imaging
• Ultrasound Technology
• Diagnostic Precision

Background:

• B-mode ultrasound images suffer from speckle noise, degrading target resolution and diagnostic accuracy.
• Speckle noise is an intrinsic limitation in ultrasound, impacting the precise diagnosis of liver conditions.

Purpose of the Study:

• To classify hepatic steatosis by reducing speckle noise in liver ultrasound tomography.
• To evaluate a novel depth attenuation correction method combined with a hybrid classification approach.

Main Methods:

• A retrospective study included 114 patients (30 normal, 44 fatty, 40 cancerous liver images).
• Depth attenuation correction was applied, followed by feature extraction from regions of interest.
• A hybrid logistic regression and support vector machine method was used for image classification with 10-fold cross-validation.

Main Results:

• The hybrid method achieved 87.5% accuracy, a 0.812 kappa statistic, and 0.119 mean absolute error.
• These results were superior to logistic regression (75.0% accuracy) and support vector machine (75.7% accuracy) alone.
• The hybrid method demonstrated improved performance, accuracy, and reduced error in classifying liver ultrasound images.

Conclusions:

• The hybrid method with depth attenuation correction offers accurate classification of normal, fatty, and cancerous liver ultrasound images.
• This approach effectively mitigates speckle noise, enhancing diagnostic capabilities.
• Future research may explore deep learning for improved liver ultrasound image classification.