Lung mass density analysis using deep neural network and lung ultrasound surface wave elastography

Area of Science:

• Medical Imaging
• Biophysics
• Machine Learning in Medicine

Background:

• Lung mass density is linked to lung pathology, but current methods like Computed Tomography (CT) do not measure it directly.
• Existing imaging techniques have limitations in directly assessing lung density and related pathologies.

Purpose of the Study:

• To develop a novel method for analyzing superficial lung tissue mass density.
• To utilize lung ultrasound surface wave elastography (LUSWE) and deep neural networks (DNNs) for this analysis.

Main Methods:

• A deep neural network (DNN) was trained using a large synthetic dataset (788,000 samples) of LUSWE measurements, including wave speed, frequency, lung mass density, and viscoelasticity.
• The DNN model, featuring 3 hidden layers with 1024 neurons each, was trained for 10 epochs.
• Validation involved a test dataset of 4000 wave speed measurements at various frequencies and experimental data from a sponge phantom.

Main Results:

• The DNN model achieved high accuracy in predicting lung mass density, with a validation accuracy of 0.992.
• Predictions from the DNN model closely matched the test dataset and experimental results from the sponge phantom.
• The LUSWE technique demonstrated effectiveness in measuring superficial lung tissue properties.

Conclusions:

• The developed method, integrating DNNs with LUSWE, shows promise for accurate lung mass density analysis.
• This technique may provide a valuable tool for assessing lung stiffness and density non-invasively.
• Further research could explore clinical applications of LUSWE for lung pathology evaluation.