LUS-BP: A Simple and Effective Brightness Profile Based Representation of LUS Data and its Classification for Pediatric Pulmonary Consolidations

Pneumonia remains one of the leading causes of death among children worldwide. Lung ultrasound (LUS) has emerged as a safe, portable and cost-effective imaging modality for identifying lung consolidations due to pneumonia. To automate the identification process, deep learning-based (DL) approaches often require the use of large-scale models and extensive computational resources. To address these challenges, we present a computationally efficient method for identifying consolidations in pediatric LUS data. In particular, we introduced a novel data representation method that transforms a two-dimensional LUS image into a one-dimensional signal, referred to as brightness profile. This is achieved by aggregating the pixel intensities along the horizontal and vertical axes of the LUS image. The compressed representation is then fed into a lightweight Multilayer Perceptron (MLP), classifying it for the presence/absence of consolidations. Results show that the model achieves a frame-level classification performance comparable to that of a computationally complex convolutional neural network (CNN) that classifies 2D LUS frames. This makes our proposed method well suited for deployment in real-world, resource-constrained clinical environments.