AI runs smack up against a big data problem in COVID-19 diagnosis

As the COVID-19 pandemic continues, the role of artificial intelligence (AI) in global health efforts becomes increasingly crucial. However, the lack of readily accessible, unified data is hampering advancements. AI has shown potential in diagnosing COVID-19 through X-ray or computed tomography (CT) scans, providing quicker results than traditional genetic testing. This has prompted the development of deep-learning neural networks to analyze these scans and streamline the radiologist's workflow.

Success stories have emerged, particularly in China, with firms like Infervision, Alibaba, and Deepwise Technology achieving notable strides in AI diagnoses. However, the lack of data accessibility and the uniqueness of COVID-19 as a disease pose significant challenges. As Wei Xu from Tsinghua University states, the deployment of these technologies has been slow, despite their potential.

Creating an AI system that efficiently identifies COVID-19-specific scan anomalies requires extensive data and collaboration with experienced physicians. The pipeline developed by Xu's team utilizes two deep learning networks refined over many years, the ResNet-50 and UNet++, to effectively differentiate COVID-19 features from other conditions. Despite their efforts, sourcing additional data for further refinement remains a significant hurdle.