Classification of Lung Disease in Children by Using Lung Ultrasound Images and Deep Convolutional Neural Network

An Interpretable Neonatal Lung Ultrasound Feature Extraction and Lung Sliding Detection System Using Object Detectors

The objective of this study was to develop an interpretable system that could detect specific lung features in neonates. A challenging aspect of this work was that normal lungs showed the same visual features (as that of Pneumothorax (PTX)). M-mode is typically necessary to differentiate between the two cases, but its generation in clinics is time-consuming and requires expertise for interpretation, which remains limited. Therefore, our system automates M-mode generation by extracting Regions of Interest (ROIs) without human in the loop. Object detection models such as faster Region Based Convolutional Neural Network (fRCNN) and RetinaNet models were employed to detect seven common Lung Ultrasound (LUS) features. fRCNN predictions were then stored and further used to generate M-modes. Beyond static feature extraction, we used a Hough transform based statistical method to detect "lung sliding" in these M-modes. Results showed that fRCNN achieved a greater mean Average Precision (mAP) of 86.57% (Intersection-over-Union (IoU) = 0.2) than RetinaNet, which only displayed a mAP of 61.15%. The calculated accuracy for the generated RoIs was 97.59% for Normal videos and 96.37% for PTX videos. Using this system, we successfully classified 5 PTX and 6 Normal video cases with 100% accuracy. Automating the process of detecting seven prominent LUS features addresses the time-consuming manual evaluation of Lung ultrasound in a fast paced environment. Clinical impact: Our research work provides a significant clinical impact as it provides a more accurate and efficient method for diagnosing lung diseases in neonates.

Machine learning-aided detection of heart failure (LVEF ≤ 49%) by using ballistocardiography and respiratory effort signals

Purpose: Under the influence of COVID-19 and the in-hospital cost, the in-home detection of cardiovascular disease with smart sensing devices is becoming more popular recently. In the presence of the qualified signals, ballistocardiography (BCG) can not only reflect the cardiac mechanical movements, but also detect the HF in a non-contact manner. However, for the potential HF patients, the additional quality assessment with ECG-aided requires more procedures and brings the inconvenience to their in-home HF diagnosis. To enable the HF detection in many real applications, we proposed a machine learning-aided scheme for the HF detection in this paper, where the BCG signals recorded from the force sensor were employed without the heartbeat location, and the respiratory effort signals separated from force sensors provided more HF features due to the connection between the heart and the lung systems. Finally, the effectiveness of the proposed HF detection scheme was verified in comparative experiments. Methods: First, a piezoelectric sensor was used to record a signal sequences of the two-dimensional vital sign, which includes the BCG and the respiratory effort. Then, the linear and the non-linear features w.r.t. BCG and respiratory effort signals were extracted to serve the HF detection. Finally, the improved HF detection performance was verified through the LOO and the LOSO cross-validation settings with different machine learning classifiers. Results: The proposed machine learning-aided scheme achieved the robust performance in the HF detection by using 4 different classifiers, and yielded an accuracy of 94.97% and 87.00% in the LOO and the LOSO experiments, respectively. In addition, experimental results demonstrated that the designed respiratory and cardiopulmonary features are beneficial to the HF detection (LVEF ≤ 49 % ). Conclusion: This study proposed a machine learning-aided HF diagnostic scheme. Experimental results demonstrated that the proposed scheme can fully exploit the relationship between the heart and the lung systems to potentially improve the in-home HF detection performance by using both the BCG, the respiratory and the cardiopulmonary-related features.