Automatic deep learning-based pleural effusion classification in lung ultrasound images for respiratory pathology diagnosis

Lung ultrasound (LUS) imaging as a point-of-care diagnostic tool for lung pathologies has been proven superior to X-ray and comparable CT, enabling earlier more accurate diagnosis in real-time at the patient’s bedside. The main limitation widespread use is its dependence on operator training experience. COVID-19 findings predominantly reflect pneumonitis pattern, with pleural effusion being infrequent. However, easy detect quantify, therefore it was selected subject of this study, which aims develop an automated system interpretation LUS effusion. A dataset collected Royal Melbourne Hospital consisted 623 videos containing 99,209 2D images 70 patients using phased array transducer. standardized protocol followed that involved scanning six anatomical regions providing complete coverage lungs respiratory pathology. This combined deep learning algorithm Spatial Transformer Network provides basis automatic pathology classification image-based level. In work, model trained supervised weakly approaches used frame- video-based ground truth labels respectively. reference expert clinician image interpretation. Both show accuracy scores test set 92.4% 91.1%, respectively, not statistically significantly different. labelling approach requires less effort from clinical experts labelling.