A Comparison of Deep Learning Methods for Liver Disease Diagnosis from Exhaled Human Breath

Abstract

The liver is a vital organ in the human body. The prevalence of liver problems has surged worldwide at an unprecedented rate as a result of unhealthy lifestyles and excessive alcohol usage. Chronic liver disease is a leading cause of mortality that impacts a significant section of the global population. Obesity, undiscovered hepatitis infection, alcohol misuse, hemoptysis or hematemesis, renal or hepatic failure, jaundice, hepatic encephalopathy, and various other conditions contribute to this condition. Therefore, prompt action is necessary to diagnose the ailment before it becomes critical. The assigned task examines several deep-learning models for gathering data from exhaled breath samples. The model's performance is evaluated based on various criteria, including accuracy, specificity, sensitivity, precision, recall, and F1-Score. These factors are examined using the training dataset to determine the training and testing loss. The proposed work does a comprehensive experimental examination of these parameters, exploring their impact on accuracy and loss function.
Additionally, it evaluates the appropriateness of these models. The deep learning models utilized in the recommended work are BiLSTM, LSTM, GRU, and 1D-CNN. The dataset is divided into 80% for training and 20% for testing, using 24 liver patient samples and 15 healthy person samples. Of the several algorithms, the BiLSTM and 1D-CNN had superior performance in predicting liver illness, achieving accuracies of 0.99 and 0.98, respectively. In addition, these two algorithms demonstrated superior precision, F1-Score, recall, specificity, and sensitivity. Therefore, these two algorithms are regarded as the superior methods for early detection of liver disease..