Applying Deep Learning Methods to Non-Alcoholic Fatty Liver Disease Management

The goal of this project is to improve computer-aided diagnosis methods for non-alcoholic fatty liver disease (NAFLD) by applying deep learning techniques. Using mostly laboratory biomarker data from NHANES 2017–18 and a b-mode ultrasound image dataset including parasagittal liver-kidney ultrasounds of 55 patients., two main non-invasive diagnostic techniques are investigated. A deep learning stacked ensemble model using a 3-layer MLP., a 5-layer MLP., and Random Forest models has been proposed for NHANES 2017–18 data. Each model's performance is evaluated separately., highlighting the benefits of the group as a whole. The liver area extraction procedure is automated for ultrasound imaging data using a refined version of a transformer model that has already been trained (U-net).). Furthermore., our technique includes taking texture characteristics out of image sequences and stacking them so that they may be used as a CNN+Bi-LSTM model's auxiliary input. AUC of 0.997., precision of 99.8%., accuracy of 99.8%., and recall of 0.997 are all attained by the stacked ensemble. The suggested approach yields 87% accuracy., 0.84 recall., 0.9 precision., 0.88 fl-score., and 0.7 specificity.