Convolutional automatic identification of B-lines and interstitial syndrome in lung ultrasound images using pre-trained neural networks with feature fusion

IntroductionInterstitial/alveolar syndrome (IS) is a condition detectable on lung ultrasound (LUS) that indicates underlying pulmonary or cardiac diseases associated with significant morbidity and increased mortality rates. However, diagnosing IS using LUS can be challenging and time-consuming, and it requires clinical expertise.MethodsIn this study, multiple convolutional neural network (CNN) models were trained as binary classifiers to accurately screen for IS in LUS frames by distinguishing between IS-present and healthy cases. The CNN models were initially pre-trained using a generic image dataset to learn general visual features (ImageNet) and then fine-tuned on our specific dataset of 108 LUS clips from 54 patients (27 healthy and 27 with IS, two clips per patient) to perform a binary classification task. Each clip in the dataset was assessed by a clinical sonographer to determine the presence of IS features or confirm healthy lung status. The dataset was split into training (70%), validation (15%), and testing (15%) sets.ResultsFollowing the process of fine-tuning, we successfully extracted features from pre-trained DL models. These extracted features were then utilised to train multiple machine learning (ML) classifiers, resulting in significantly improved accuracy in IS classification compared with the individual CNN models. Advanced visual interpretation techniques such as heatmaps based on gradient-weighted class activation mapping (Grad-CAM) and local interpretable model-agnostic explanations (LIME) were implemented to further analyse the outcomes. The best-trained ML model achieved a test accuracy rate of 98.2%, with specificity, recall, precision, and F1 score values above 97.9%.ConclusionOur study demonstrates the feasibility of using a pre-trained CNN as a diagnostic tool for IS screening on LUS frames, integrating targeted data filtering, feature extraction, and fusion techniques. The data-filtering technique refines the training dataset by excluding LUS frames that lack IS-related features (e.g., absence of B-lines). Feature fusion combines features learnt from different models or “fused” to enhance overall predictive performance. This study confirms the practicality of using pre-trained CNN models with feature extraction and fusion techniques for screening IS using LUS frames. This represents a noteworthy advancement in improving the efficiency of diagnosis. In the next steps, validation on larger datasets will assess the applicability and robustness of these CNN models in more complex clinical settings.