Liver Fat Assessment with Body Shape.

Area of Science:

• Medical imaging and artificial intelligence
• Non-invasive diagnostic techniques
• Obesity and metabolic disease research

Background:

• Hepatic steatosis (fatty liver disease) is a growing health concern, particularly in obese individuals, increasing risks for cirrhosis and liver cancer.
• Current diagnostic methods like liver biopsy and CT/MR imaging are invasive, costly, or carry health risks.
• There is a need for accessible and affordable methods to monitor liver fat content.

Purpose of the Study:

• To investigate the utility of body shape features for assessing hepatic steatosis.
• To develop and compare traditional linear regression and deep neural network models for predicting liver fat.
• To evaluate the performance of these models in both regression and classification tasks.

Main Methods:

• Utilized a medical dataset containing body shape measurements.
• Developed and applied traditional linear regression models.
• Implemented and trained a deep neural network model for hepatic steatosis assessment.
• Compared model performance using metrics such as Root Mean Square Error (RMSE) and accuracy.

Main Results:

• The deep neural network model significantly outperformed the linear regression model.
• The neural network achieved a 22.37% improvement in RMSE and an 18% increase in accuracy compared to linear regression.
• The neural network model demonstrated an R-squared value exceeding 0.72 and an accuracy of 78%.

Conclusions:

• Body shape features offer a valuable, accurate, and affordable supplementary method for monitoring liver fat.
• The developed deep neural network approach shows high potential for clinical application and telehealth, especially with the proliferation of depth-sensing cameras.
• While not replacing gold-standard diagnostics, this method provides a convenient tool for widespread use in clinical and home settings.