Applying Radiomics and Deep Learning to Investigations into Seafarers' Health Status

Abstract

Objective This study aimed to conduct a retrospective investigation into the prevalence of fatty liver disease and gallstones among seafarers, explore the relevant influencing factors, and further evaluate the application value of deep learning-based radiomics in intelligent screening, quantitative diagnosis, and risk prediction of these hepatobiliary diseases.   Methods A total of 7232 male seafarers who underwent abdominal ultrasound (Apollo 500/Esaote MyLab Gamma system), blood lipid testing (Hitachi 7180 automatic biochemical analyzer), and body mass index (BMI) measurement at the Physical Examination Center of Shanghai Waterway Hospital from June 2022 to November 2023 were randomly selected as research subjects. Traditional statistical methods (chi-squared test, correlation analysis, t-test) were used to analyze the correlation between clinical indicators (BMI, blood lipids) and disease incidence. Meanwhile, a deep learning framework integrating U-Net-based image segmentation and MobileNetV2-based classification was constructed: 1) Automatically segmenting the liver and gallbladder regions of interest (ROIs) from ultrasound images to avoid subjective errors of manual segmentation; 2) Extracting high-dimensional radiomic features (texture, gray level, morphological features) through the pre-trained deep learning model, replacing traditional manual feature selection; 3) Establishing a dual disease prediction model (fatty liver/gallstones) by fusing clinical indicators (BMI, blood lipids) and deep radiomic features.   Results With the increase in BMI, the incidence rates of hyperlipidemia and fatty liver disease showed a linear positive correlation (r=0.98, t=6.83, df=2, P<0.05). The deep learning model achieved excellent performance in disease diagnosis: for fatty liver disease, the area under the receiver operating characteristic curve (AUC) was 0.93, accuracy 90.2%, recall 88.5%, and specificity 89.8%, which was significantly higher than traditional manual ultrasound diagnosis (AUC=0.79, P<0.05); for gallstones, the model achieved AUC=0.89, accuracy 87.6%, and recall 85.3%. Chi-squared test confirmed that gallstone formation was statistically correlated with thickened and rough gallbladder walls, hyperlipidemia, and hypercholesterolemia (P<0.005).   Conclusion Controlling BMI and reducing blood lipid levels can effectively lower the incidence of fatty liver disease. Deep learning-based radiomics realizes "automated segmentation-quantitative feature extraction-intelligent prediction" of hepatobiliary diseases, which is particularly suitable for seafarers with limited on-board medical resources and large-scale physical examination scenarios. Popularizing this AI tool, combined with targeted health education and lifestyle interventions, will significantly improve the efficiency and accuracy of seafarers' hepatobiliary health management.