Multimodal Convolutional Neural Network Model for Evaluating Development of Fetal Rabbit Lung Using B-mode and Shear...

Area of Science:

• Medical imaging
• Fetal development
• Artificial intelligence in medicine

Background:

• Assessing fetal lung maturity is crucial for determining optimal delivery timing.
• Current methods for evaluating fetal lung development can be invasive or lack precision.

Purpose of the Study:

• To evaluate convolutional neural network (CNN)-based models for predicting fetal rabbit lung development.
• To compare unimodal models (B-mode and shear wave elastography - SWE) with a multimodal model (B-mode + SWE).

Main Methods:

• Acquired 1670 ultrasound images (B-mode and SWE) from 167 fetal rabbits (23-30 d gestation).
• Developed CNN models based on fetal lung histological classification (canalicular, saccular, alveolar stages).
• Extracted features from B-mode or SWE images to predict fetal lung development stages.

Main Results:

• The SWE unimodal model achieved superior overall accuracy (95.9%) compared to B-mode (86.5%) and multimodal (87.9%) models.
• SWE models outperformed B-mode in predicting canalicular and saccular lung development stages.
• For the alveolar stage, SWE and multimodal models showed higher accuracy, specificity, PPV, and AUC than B-mode, with the multimodal model having a slight edge in specificity and PPV.

Conclusions:

• SWE-based models demonstrate high performance in predicting fetal rabbit lung development.
• These models offer a promising non-invasive approach for clinical assessment of fetal lung maturity.
• SWE imaging combined with CNNs represents a significant advancement in prenatal diagnostics.