S2FLNet: Hepatic steatosis detection network with body shape.

Area of Science:

• Medical imaging and diagnostics
• Artificial intelligence in healthcare
• Cardiovascular disease research

Background:

• Hepatic steatosis (fatty liver disease) increases cardiac complication and cardiovascular mortality risks.
• Current detection methods like biopsy, MRI, and CT scans are costly and may involve medical complications.
• Accurate and accessible detection of hepatic steatosis is critically important for patient outcomes.

Purpose of the Study:

• To propose a novel deep neural network for estimating hepatic steatosis degree using only body shape information.
• To develop a non-invasive and accessible method for assessing fatty liver disease.
• To improve upon the accuracy and accessibility of current hepatic steatosis detection techniques.

Main Methods:

• Utilized a deep neural network incorporating dilated residual network blocks to extract refined body shape features.
• Implemented a hybrid loss function combining center loss and cross-entropy loss for improved classification accuracy.
• Trained and tested the network on a public medical dataset to evaluate performance.

Main Results:

• The proposed deep neural network achieved a total accuracy exceeding 82% in estimating hepatic steatosis degree (low, mid, high).
• The network demonstrated effective feature extraction and classification capabilities using body shape data.
• Experimental results validated the network's performance across various parameters.

Conclusions:

• The developed deep neural network provides an accurate and accessible method for assessing hepatic steatosis.
• This AI-driven approach offers a potential non-invasive alternative to traditional diagnostic methods.
• Further research can explore the clinical application of this body shape-based AI for fatty liver disease detection.