FPT Software has recently developed an AI-powered model that helps detect pneumothorax through image segmentation. The model is also applicable in the early detection of lung diseases such as pneumonia, sarcoidosis, cystic fibrosis, tuberculosis, lung cancer, and idiopathic pulmonary fibrosis.
In 2019, FPT Software developed a network architecture called Residual Capsule Networks for medical image segmentation, based on which it built a deep-learning model to identify pneumothorax areas on the medical X-ray images.
From more than 10,000 2D chest X-ray images labelled by doctors from all over the world, the Residual Capsule Network-based AI model could assess, classify, and segment images that show areas of a high risk of pneumothorax, thus improving the diagnosis performance. According to FPT Software Head of Digital Innovation Group Quang Tran, generated results have shown an impressive accuracy of 83.92 per cent.
“This model could detect a respiratory infection that affects lungs and other internal organs”, he said. “The key challenge is to collect data from new infected cases so that we could train and upgrade the model to carry out a diagnosis from radiology, instead of conventional sputum tests”.
“Amid increasing respiratory diseases due to air pollution and global warming, we believe that AI technologies have great potential in healthcare, particularly in diagnosis and treatment of diseases”, Quang noted.
FPT Software’s senior AI scientist said that the company is willing to collaborate with domestic and overseas hospitals to collect data and further develop the model, helping to diagnose a wider range of lung diseases fast and accurately.
Being a global IT and digital transformation services provider, FPT Software has been active in research and development of next-gen technologies such as AI and IoT with the hope to revolutionize healthcare. Earlier in January, the company’s AI-powered algorithm, Residual Capsule, competed against more than 1,400 worldwide competitors to rank top 5 per cent at the Pneumothorax Challenge, co-hosted by The Society for Informatics in Medicine (SIIM) and The American College of Rheumatology (ACR).
FPT Software has recently registered the patent for Residual Capsule Networks for Medical Image Segmentation. For more information about Residual Capsule Networks for Medical Image Segmentation, please contact Nu Phan (Mrs) – [email protected].