Preliminary application of AI diagnosis system in the diagnosis of the novel coronavirus infected pneumonia(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2020年第12期
- Page:
- 1604-1608
- Research Field:
- 医学人工智能
- Publishing date:
Info
- Title:
- Preliminary application of AI diagnosis system in the diagnosis of the novel coronavirus infected pneumonia
- Author(s):
- DENG Lingbo1; ZHOU Wen1; ZHAO Shuangquan2; XIANG Ziyun3; CHENG Guanxun1
- 1. Department of Medical Imaging, Shenzhen Hospital of Peking University, Shenzhen 518036, China 2. Department of Radiology, Baoan Peoples Hospital, Shenzhen 518101, China 3. Department of Radiology, Longgang Peoples Hospital, Shenzhen 518172, China
- Keywords:
- Keywords: novel coronavirus infected pneumonia computed tomography artificial intelligence diagnosis system
- PACS:
- R318；R563.1
- DOI:
- DOI:10.3969/j.issn.1005-202X.2020.12.025
- Abstract:
- Abstract: Objective To evaluate the diagnostic value of artificial intelligence (AI) assisted diagnosis system for new coronavirus pneumonia (NCIP). Methods The clinical data and CT images of 26 NCIP patients were retrospectively analyzed. The new coronary pneumonia diagnosis module of the AI system is used to automatically identify the number of pneumonia lesions, measure the volume of the lesion, and calculate the percentage of the lesion volume in the lung lobe to obtain the probability of suspected new coronary pneumonia. Evaluate the accuracy of the AI system in identifying the number of lesions and the accuracy of the scope of the lesion. Results The AI system automatically detected 215 lesions of 26 NCIP patients. In 76.9% (20/26) of the patients AI has recognized more lesions than the doctor did. 23.1% (6/26) of the patients had a lesion volume less than 10 cm3, 38.5% (10/26) had a lesion volume of 10-100 cm3, and 38.5% (10/26) had a lesion volume greater than 100 cm3. 57.7% (15/26) patients have lesion volume percentage less than 10%, 23.1% (6/26) patients have a volume percentage of 10%~25%, 15.4% (4/26) patients have a volume percentage of 25%~50%, and only 3.8 % (1/26) of the patient have a volume percentage greater than 50%. 34.6% (9/26) patients have a suspected probability of NCIP greater than 50%, of which only 11.5% (3/26) patients have a suspected probability of 99%, and 65.4% (17/26) patients have a suspected probability of less than 25%. 61.5% (16/26) of the patients have some smaller lesions unrecognized by AI, 38.5% (10/26) of the patients have two adjacent lesions that was mistakenly recognized as the same lesion, and there are also 42.3% (11/26) of the patients whose lungs partial artifacts were identified as lesions, 61.5% (16/26) of the patients whose normal lung structures were identified as lesions, 30.8% (8/26) of the patients whose lungs were identified as NCIP lesions, and 46.2% (12/26) of the patients whose small lesions were identified as lung nodules. In 88.5% (23/26) patients, some lung tissues with normal edges of the lesion were recognized as lesions in 61.5% (16/26) of the patients the lesion edge was not recognized. Conclusion The AI-assisted diagnosis system can quickly identify pneumonia lesions in CT images, and measure the volume, give the suspected probability of NCIP, help doctors quickly identify high-risk groups, determine the severity and efficacy of the disease, save time and energy, and achieve precise prevention and control Effect.
Last Update: 2020-12-30