RESUMEN
Artificial intelligence can potentially provide a substantial role in streamlining chest computed tomography (CT) diagnosis of COVID-19 patients. However, several critical hurdles have impeded the development of robust AI model, which include deficiency, isolation, and heterogeneity of CT data generated from diverse institutions. These bring about lack of generalization of AI model and therefore prevent it from applications in clinical practices. To overcome this, we proposed a federated learning-based Unified CT-COVID AI Diagnostic Initiative (UCADI, http://www.ai-ct-covid.team/), a decentralized architecture where the AI model is distributed to and executed at each host institution with the data sources or client ends for training and inferencing without sharing individual patient data. Specifically, we firstly developed an initial AI CT model based on data collected from three Tongji hospitals in Wuhan. After model evaluation, we found that the initial model can identify COVID from Tongji CT test data at near radiologist-level (97.5% sensitivity) but performed worse when it was tested on COVID cases from Wuhan Union Hospital (72% sensitivity), indicating a lack of model generalization. Next, we used the publicly available UCADI framework to build a federated model which integrated COVID CT cases from the Tongji hospitals and Wuhan Union hospital (WU) without transferring the WU data. The federated model not only performed similarly on Tongji test data but improved the detection sensitivity (98%) on WU test cases. The UCADI framework will allow participants worldwide to use and contribute to the model, to deliver a real-world, globally built and validated clinic CT-COVID AI tool. This effort directly supports the United Nations Sustainable Development Goals number 3, Good Health and Well-Being, and allows sharing and transferring of knowledge to fight this devastating disease around the world.
RESUMEN
Background and purposeThe worldwide pandemic of coronavirus disease 2019 (COVID-19) greatly challenges public medical systems. With limited medical resources, the treatment priority is determined by the severity of patients. However, many mild outpatients quickly deteriorate into severe/critical stage. It is crucial to early identify them and give timely treatment for optimizing treatment strategy and reducing mortality. This study aims to establish an AI model to predict mild patients with potential malignant progression. MethodsA total of 133 consecutively mild COVID-19 patients at admission who was hospitalized in Wuhan Pulmonary Hospital from January 3 to February 13, 2020, were selected in this retrospective IRB-approved study. All mild patients were categorized into groups with or without malignant progression. The clinical and laboratory data at admission, the first CT, and the follow-up CT at the severe/critical stage of the two groups were compared. Both multivariate logistic regression and deep learning-based methods were used to build the prediction models, with their area under ROC curves (AUC) compared. ResultsMultivariate logistic regression depicted 6 risk factors for malignant progression: age >55years (OR 5.334, 95%CI 1.8-15.803), comorbid with hypertension (OR 5.093, 95%CI 1.236-20.986), a decrease of albumin (OR 4.01, 95%CI 1.216-13.223), a decrease of lymphocyte (OR 3.459, 95%CI 1.067-11.209), the progressive consolidation from CT1 to CTsevere (OR 1.235, 95%CI 1.018-1.498), and elevated HCRP (OR 1.015, 95%CI 1.002-1.029); and one protective factor: the presence of fibrosis at CT1 (OR 0.656, 95%CI 0.473-0.91). By combining the clinical data and the temporal information of the CT data, our deep learning-based models achieved the best AUC of 0.954, which outperformed logistic regression (AUC: 0.893), ConclusionsOur deep learning-based methods can identify the mild patients who are easy to deteriorate into severe/critical cases efficiently and accurately, which undoubtedly helps to optimize the treatment strategy, reduce mortality, and relieve the medical pressure.
RESUMEN
Objective To summarize CT and MRI features of hyaline vascular type localized Castleman disease(LCD)and analyze the causes of misdiagnosis,to improve the preoperative diagnosis rate.Methods The clinical and imaging data of 7 patients with hyaline vascular type LCD confirmed by operation and pathology were analyzed retrospectively.Results (1)6 cases were misdiagnosed before operation,1 case was misdiagnosed as pancreatic neuroendocrine tumor,1 case as thymoma,1 case as neurogenic tumor,1 case as pheochromocytoma, 1 case as clear cell renal cell carcinoma and 1 case as small mesenteric stromal tumor.(2)1 case was located in the right neck,1 case in the anterior superior mediastinum,1 case in the neck of the pancreas,1 case in the upper part of the left kidney,2 cases in the retroperitoneum and 1 case in the lower abdomen.(3)3 cases were scaned by dynamic enhanced MRI,3 cases were scaned by dynamic enhanced CT, and 1 case was checked by plain CT and enhanced MRI.CT and MRI showed that 7 cases had a round or elliptical soft tissue mass, and 4 cases with well defined margin,3 cases were not clear in edge,2 cases with spot or strip calcification on CT images,4 cases had slightly longer T1 and longer T2 signal,4 cases were restricted of diffusion and had higher signal on DWI.All the lesions were enhanced in arterial phase,and went on in the delayed phase.There were 5 cases with distorted vascular shadow in the middle and/or around of the mass, 3 cases with strips,spoke-like low-density areas or low-signal areas,and some lesions were filled in delayed phase.Conclusion CT and MRI features of hyaline vascular type LCD have certain characteristics such as rich blood supply,enhancement in persistent,tortuosity of peripheral vascular,with some short strip calcification and high signal on DWI,which may be helpful for preoperative diagnosis.