AI Diagnoses Lung Diseases from Ultrasound Videos with 96.57% Accuracy

The new study by researchers from Charles Darwin University (CDU, Casuarina, Australia) and other collaborating institutions aims to enhance AI's diagnostic capabilities by training it to analyze lung ultrasound videos for signs of respiratory diseases. The AI model works by examining each individual frame of the ultrasound video to identify key features of the lungs. It also analyzes the sequence of video frames to detect patterns in the lungs' behavior over time. Using this data, the model can pinpoint specific patterns indicative of various lung conditions, categorizing the ultrasound as normal, or as showing signs of pneumonia, COVID-19, or other lung diseases. In addition, the model employs AI techniques to provide explanations for its diagnostic decisions, helping radiologists understand and trust its results. The model utilizes explainable AI, a technique that makes the outputs of machine learning algorithms more transparent to users. By using visual aids like heatmaps, the system helps doctors understand why certain conclusions were made, making it easier for them to localize areas of concern and significantly improving clinical transparency.

This interpretability aims to enhance the reliability of the AI system, making it a useful tool for physicians. Not only can the model help doctors quickly and accurately diagnose lung diseases, but it also supports their decision-making process, saves valuable time, and serves as an educational resource. The findings, published in Frontiers in Computer Science, demonstrate that the model achieved an accuracy rate of 96.57%, with its analyses verified by medical professionals. The researchers believe that with proper training and data, the model could be expanded to detect additional diseases such as tuberculosis, black lung disease, asthma, cancer, chronic lung disease, and pulmonary fibrosis. Future research could involve training the AI model to assess other types of medical imaging, including CT scans and X-rays.

Related Links:
Charles Darwin University