Lianxin Healthcare and Baidu PaddlePaddle Launch Industry's First Open-Source AI Model for Pneumonia CT Image Analysis to Support Global Pandemic Response

On February 28, Defiled Medical launched the “AI System for Pneumonia Screening and Condition Pre-assessment Based on CT Images,” developed on the Baidu PaddlePaddle platform. The system has been initially deployed at the Affiliated Hospital of Xiangnan University in Chenzhou, Hunan Province.

Leveraging the core AI technologies accumulated by Defiled Medical in the field of medical imaging, and developed in conjunction with the PaddlePaddle open-source framework and the visually advanced PaddleSeg development suite, this system enables rapid detection and identification of pneumonia lesions. It provides quantitative assessment information for disease diagnosis, including lesion count, volume, and lung involvement percentage. Additionally, it employs visualization tools such as histograms of bilateral lung density distribution and overlaid lesion segmentation displays, offering clinicians both qualitative and quantitative bases for screening and preliminary diagnosis of pneumonia, thereby enhancing diagnostic and evaluation efficiency.

CT imaging has become a critical basis for screening and diagnosing COVID-19. During this crucial period of pandemic management, the total number of existing and new patients is substantial. Physicians need to conduct follow-up comparisons of multiple CT scans taken at different stages of disease progression to accurately assess patient condition development and treatment efficacy. If traditional visual inspection methods are used, physicians not only face an enormous workload but also struggle to achieve precise assessment and timely comparison of patient conditions. Amidst the nationwide anti-pandemic effort, where medical resources are scarce and doctors are overworked, the excessive volume of CT examinations undoubtedly poses a significant challenge to medical resource allocation on the front lines, thereby affecting the speed of patient diagnosis and treatment. The launch of an AI-based CT imaging system for pneumonia screening and preliminary condition assessment can effectively help alleviate clinical physicians' work pressure, accelerate patient diagnosis and treatment, and contribute to mitigating medical resource shortages and achieving ultimate victory in the fight against the pandemic.

(Figure: Baidu PaddlePaddle assists Defiled Medical in developing an AI system for pneumonia screening and condition pre-assessment based on CT images)

The AI-based pneumonia screening and preliminary assessment system, first deployed at the Affiliated Hospital of Xiangnan University, can complete a full suite of quantitative analyses and visualizations within tens of seconds. These include lesion detection, lesion contour delineation, histogram analysis of bilateral lung density distribution, as well as calculation and display of metrics such as the number, volume, and lung parenchymal involvement ratio of pulmonary lesions. On the test dataset, the system achieved a lesion detection precision of 92% and a recall rate of 97%, thereby ensuring high detection accuracy while minimizing missed diagnoses. During the system’s development, the Affiliated Hospital of Xiangnan University established rigorous annotation standards for algorithm training data, set strict acceptance criteria for model deployment, and provided professional clinical guidance for the design of functional modules.

Particularly noteworthy is that the deep learning algorithm model employed by this system has been extensively trained on collected high-resolution and low-resolution CT imaging data, enabling it to adapt exceptionally well to examination data acquired from CT scanners of varying tiers. This system holds promise for providing an effective AI-assisted preliminary diagnostic tool for pneumonia at primary healthcare institutions with limited medical resources and capabilities.

The system will subsequently be deployed in hospitals across Hubei, Chengdu, and other regions to assist local physicians in combating the epidemic. Leveraging Baidu’s PaddlePaddle deep learning platform, the online version of the system will be offered free of charge to designated hospitals nationwide. This initiative facilitates remote consultation and collaboration among medical professionals, enhances diagnostic and treatment capabilities at primary healthcare institutions, and is expected to reduce the risk of cross-infection during patient referrals and mobile medical visits.

Driven by a shared commitment to corporate social responsibility and a consensus on leveraging technology for societal benefit, Baidu PaddlePaddle has supported Lianxin Medical in open-sourcing the AI model for pneumonia CT image analysis from its system—the first of its kind in the industry. Released on the Baidu PaddlePaddle platform, this tool serves as a powerful resource for research and development across the sector. We look forward to greater participation from hospitals and algorithm researchers in the development of AI-based big data medical imaging products for epidemic control, thereby contributing to clinical research and product development in the fight against the pandemic.

Meanwhile, this model is also available as a pre-trained model through the EasyDL custom training and service platform. Developers can select the “Pneumonia CT Image Recognition Specialized Algorithm” within EasyDL’s image segmentation models, train with a small amount of data to obtain a model further optimized for real-world scenarios, and access ready-to-use model services that flexibly support various deployment options.

Beijing Defiled Medical Technology Co., Ltd. is a domestic leader in intelligent medical imaging research. Leveraging artificial intelligence and cloud computing technologies, the company is dedicated to providing hospitals and third-party imaging centers with technical tools and cloud services for the automatic segmentation of organs and lesions. During this critical period of nationwide epidemic prevention and control, Defiled Medical swiftly leveraged its accumulated expertise to rapidly develop, launch, and open-source a CT-based pneumonia screening and disease severity pre-assessment system on the PaddlePaddle platform, striving to contribute its full strength to the fight against the epidemic.

As China’s only open-source, open, and fully-featured industrial-grade deep learning platform, PaddlePaddle has long been committed to open-source initiatives. During the pandemic, PaddlePaddle freely released the industry’s first open-source model for mask face detection and classification, and supported Beijing Defiled Medical Technology Co., Ltd. in developing and open-sourcing an AI model for CT image analysis of pneumonia. By leveraging Baidu’s leading AI capabilities, PaddlePaddle provided rapid and effective support to businesses, government agencies, and other organizations involved in the joint fight against the epidemic.