AI-Powered CT Quantitative Analysis Accurately Predicts Clinical Classification of COVID-19 Pneumonia

The global COVID-19 pandemic has posed an unprecedented challenge to clinical diagnosis and treatment of infectious diseases, with timely and accurate assessment of disease severity being critical to improving patient prognosis and optimizing clinical intervention strategies. While reverse transcription-polymerase chain reaction (RT-PCR) remains the gold standard for COVID-19 diagnosis, chest computed tomography (CT) has emerged as an indispensable auxiliary tool due to its rapidity, high sensitivity in detecting pulmonary lesions, and ability to visualize the extent of lung involvement. In recent years, the integration of artificial intelligence (AI) into medical imaging has revolutionized the quantitative analysis of CT scans, moving beyond traditional qualitative and semi-quantitative evaluations to provide objective, precise, and reproducible data for clinical decision-making. A groundbreaking study published in CT Theory and Applications has demonstrated that AI-based CT quantitative analysis of key pulmonary lesion metrics—including total lung infection volume, ground-glass opacity (GGO) volume, and solid opacity (SO) volume—exhibits a strong correlation with the clinical classification of COVID-19 pneumonia, offering a reliable objective imaging basis for clinicians to quickly assess disease severity and implement targeted treatment plans.

Against the backdrop of the COVID-19 pandemic, the clinical classification of COVID-19 pneumonia—into common, severe, and critical types—has become a cornerstone of clinical management, guiding the allocation of medical resources, the selection of treatment regimens, and the prediction of patient outcomes. However, traditional CT image analysis relies on the subjective judgment of radiologists, which may be affected by factors such as clinical experience, inter-observer variability, and the complexity of pulmonary lesions, leading to inconsistencies in the assessment of disease severity. In addition, the rapid progression of COVID-19 and the surge in the number of patients have placed enormous pressure on clinical radiology departments, making it difficult for manual image analysis to meet the needs of rapid diagnosis and real-time disease monitoring. The application of AI technology in CT quantitative analysis has addressed these pain points: AI diagnostic systems can automatically identify, segment, and quantify pulmonary lesions in a short time, eliminate human subjective bias, and provide standardized quantitative indicators, which is of great significance for improving the efficiency and accuracy of clinical diagnosis and treatment of COVID-19 pneumonia.

The study, a retrospective clinical research project, was conducted by a research team led by Liu Li from the CRT Clinical R&D Group of The First Hospital of Qiqihar, in collaboration with Li Huixin from Hangzhou Yitu Medical Technology Co., Ltd., and Gao Xiaolong from the School of Medical Technology of Qiqihar Medical University. The research team aimed to systematically evaluate the correlation between AI-derived CT quantitative indicators and the clinical classification of COVID-19 pneumonia, and to verify the clinical value of AI-powered CT quantitative analysis in predicting and assessing the severity of COVID-19 pneumonia. To ensure the scientificity and rigor of the research, the team strictly formulated inclusion and exclusion criteria for research subjects, retrospectively collecting clinical and chest CT imaging data of 46 confirmed COVID-19 patients admitted to the fever clinic of The First Hospital of Qiqihar from February 1, 2020 to January 20, 2021. All included patients had positive results from multiple pharyngeal swab RT-PCR tests and met the diagnostic criteria for common, severe, or critical COVID-19 pneumonia specified in the Diagnostic and Treatment Protocol for Novel Coronavirus Pneumonia (Trial Eighth Edition) issued by the National Health Commission of the People’s Republic of China. Patients with other known viral pneumonia, mycoplasma or chlamydia pneumonia, and those with poor-quality chest CT images were excluded from the study, effectively controlling confounding factors and ensuring the homogeneity of the research population.

The demographic characteristics of the 46 included patients showed a balanced gender distribution, with 23 males and 23 females, and an average age of 43.85 ± 16.29 years, ranging from 12 to 71 years old. The research team collected detailed clinical symptoms of all patients, including fever, cough, chest tightness, dyspnea, headache, sore throat, myalgia, and diarrhea, and classified the patients into three clinical subtypes according to the national diagnostic criteria: 26 cases of common type, 16 cases of severe type, and 4 cases of critical type. A comprehensive statistical analysis of the clinical data revealed that fever, cough, and fatigue were the main clinical manifestations of the included patients, with the incidence of fever reaching nearly 98%, reflecting the typical systemic and respiratory symptoms of COVID-19 pneumonia. Notably, there were no statistically significant differences in gender distribution and the incidence of major clinical symptoms among the three clinical subtypes, indicating that these clinical features alone are not sufficient to distinguish the severity of COVID-19 pneumonia. In contrast, age showed a significant correlation with clinical classification: the average age of patients with common type was 37.8 ± 14.1 years, while that of severe and critical type patients was 51.4 ± 13 years and 58.5 ± 9.9 years, respectively. Statistical tests confirmed that the age of common type patients was significantly lower than that of severe and critical type patients, and there was no significant age difference between severe and critical type patients. This finding suggests that advanced age is an important risk factor for the progression of COVID-19 pneumonia to severe and critical types, which is consistent with the clinical epidemiological characteristics of COVID-19 reported in previous global studies, and provides an important clinical reference for identifying high-risk groups of severe COVID-19 pneumonia.

In the CT examination part of the study, the research team adopted a standardized scanning protocol to ensure the consistency and comparability of CT image data. All patients underwent chest CT scanning using a GE Optima 660 64-slice CT scanner, with the patient in the supine position and scanning performed during breath-holding after inspiration, covering the range from the lung apex to the lung base. The scanning parameters were set as follows: tube voltage 80-120 kV, tube current 50-350 mAs, pitch 0.99-1.22 mm, matrix 512×512, slice thickness 10 mm, and field of view 350 mm×350 mm. Multiplanar reconstruction (MPR) technology was used for post-processing of the scanning data, with a reconstructed slice thickness of 0.625 mm. The lung window and mediastinal window were set to standard parameters: lung window with a window width of 1200 HU and a window level of -600 HU, and mediastinal window with a window width of 350 HU and a window level of 40 HU. All CT scanning data were transmitted to the picture archiving and communication system (PACS), and the AI intelligent auxiliary diagnosis system independently developed by Hangzhou Yitu Medical Technology Co., Ltd. was loaded for quantitative analysis of pulmonary lesions. This standardized scanning and post-processing protocol effectively reduced the technical variability in CT image acquisition and processing, laying a solid foundation for the accuracy and reliability of subsequent AI quantitative analysis.

The core of the study is the quantitative analysis of COVID-19 pneumonia lesions using the AI auxiliary diagnosis system, which realized the automatic identification, segmentation and volume calculation of total lung infection lesions, GGO lesions and SO lesions. Two senior radiologists independently used the AI system software to analyze and evaluate the pulmonary lesions of all patients, and the consistency of the evaluation results was verified to ensure the credibility of the AI quantitative data. The AI system uses advanced deep learning algorithms and medical image segmentation technology to accurately identify the boundary of COVID-19 pneumonia lesions in CT images, distinguish GGO and SO lesions according to the CT value of the lesions, and calculate the volume of various lesions in cubic centimeters, which overcomes the limitations of traditional manual measurement that is time-consuming, laborious and prone to errors. The statistical analysis of the AI quantitative data showed that the CT manifestations of pulmonary lesions in all three clinical subtypes of COVID-19 pneumonia were dominated by GGO, which is a typical early imaging feature of COVID-19 pneumonia, reflecting the pathological changes of alveolar edema, exudation and interstitial inflammation in the early stage of the disease. There were significant differences in the total lung infection volume, GGO volume and SO volume among the three clinical subtypes: the quantitative indicators of the common type were significantly lower than those of the severe and critical types, while there was no statistically significant difference in the three quantitative indicators between the severe and critical types. Specifically, the total lung infection volume of common type patients was 48.6 (21.2, 129.6) cm³, GGO volume was 27.1 (10.5, 58.0) cm³, and SO volume was 25.7 (7.6, 64.4) cm³; the corresponding indicators of severe type patients were 588.9 (474.0, 927.9) cm³, 338.4 (274.4, 563.8) cm³ and 255.9 (189.8, 364.0) cm³; and the critical type patients had total lung infection volume of 1065.4 (826.9, 1183.6) cm³, GGO volume of 657.2 (637.3, 811.9) cm³ and SO volume of 301.9 (174.5, 493.1) cm³. Kruskal-Wallis rank sum test confirmed that the differences in the three quantitative indicators among the three groups were statistically significant (P < 0.001), and pairwise comparison using the least significant difference (LSD) method showed that the differences between the common type and the severe/critical type were statistically significant (P 0.05).

To further explore the relationship between AI-based CT quantitative indicators and the clinical classification of COVID-19 pneumonia, the research team conducted a Spearman rank correlation analysis, and the results showed that the total lung infection volume, GGO volume and SO volume all had a strong positive correlation with the clinical classification of COVID-19 pneumonia, with correlation coefficients of 0.863, 0.870 and 0.841, respectively (all P < 0.001). This high correlation coefficient fully demonstrates that the larger the total lung infection volume, GGO volume and SO volume, the more severe the clinical classification of COVID-19 pneumonia, which means that these AI-derived CT quantitative indicators can effectively reflect the severity of lung tissue damage caused by COVID-19 infection. From a pathological perspective, the high correlation between the quantitative indicators and clinical classification is consistent with the pathological development process of COVID-19 pneumonia: GGO is the early pathological manifestation of COVID-19 pneumonia, which is caused by alveolar swelling, alveolar cavity exudation and interstitial inflammatory response, leading to an increase in the CT value of pulmonary lesions; with the progression of the disease, the inflammatory exudation in the alveoli and interstitium increases, the CT value of the lesions further rises and exceeds that of the intrapulmonary blood vessels, and the lesions transform from GGO to SO, which is accompanied by the expansion of the lesion range and the increase of the total infection volume, and the patient's clinical condition gradually progresses from common type to severe and critical type. Therefore, the AI-based CT quantitative analysis of GGO and SO volume not only reflects the extent of lung tissue involvement, but also indirectly reflects the pathological stage of COVID-19 pneumonia, providing a bridge between imaging manifestations and pathological changes for clinical diagnosis and treatment.

The study also deeply discussed the clinical advantages of AI-powered CT quantitative analysis in the assessment of COVID-19 pneumonia clinical classification. The AI diagnostic system developed by Hangzhou Yitu Medical Technology Co., Ltd. can not only calculate the volume of various pulmonary lesions, but also automatically identify the distribution location of infected lesions, calculate the ratio of infected volume to normal lung tissue, and judge the type of infected lesions according to the CT threshold, and even realize the preliminary classification of the clinical severity of COVID-19 pneumonia. For mild COVID-19 patients, the early CT manifestations are small ground-glass opacities and interstitial changes, which are easy to be missed by traditional manual CT observation, while the AI system has high sensitivity and can quickly identify and evaluate these subtle lesions; for common type patients, the CT manifestations are mostly single or bilateral multiple ground-glass opacities with limited lesion range and uneven density, and the AI system can accurately quantify the lesion volume and provide objective data for clinical judgment; for severe and critical type patients, the CT lesions are widely distributed, involving multiple lung lobes or segments with asymmetric distribution, and the AI system can quickly complete the quantitative analysis of extensive lesions in a short time, which is far more efficient than manual analysis. In the clinical practice of the COVID-19 pandemic, especially during the peak of the epidemic, the AI system can significantly improve the work efficiency of radiology departments, reduce the work burden of radiologists, and realize the rapid screening and severity assessment of a large number of suspected COVID-19 patients, which is of great significance for the rapid control of the epidemic.

In addition, the study also combined the existing research results to discuss the clinical significance of GGO and SO volume in the evaluation of COVID-19 pneumonia severity. Previous studies have shown that alveolar injury exudation is the main pathological change of the lungs in COVID-19 patients, so SO volume is often regarded as an important indicator to evaluate the severity of COVID-19 patients. The study found that the pulmonary lesions of all included patients were dominated by GGO, which is not contradictory to the above conclusion, because all patients underwent the first chest CT examination within 2 days of admission, and GGO is the typical early CT sign of COVID-19 pulmonary infection. With the progression of the disease, GGO lesions will gradually transform into SO lesions, so GGO is the early pathological manifestation of SO lesions. This finding further clarifies the dynamic relationship between GGO and SO in the development of COVID-19 pneumonia, and indicates that the AI-based quantitative analysis of GGO volume in the early stage of the disease can predict the potential progression trend of the disease, providing an important basis for clinical early intervention of high-risk patients.

While the study has achieved important clinical results, the research team also objectively pointed out the limitations of the study, which provides a clear direction for the further research and optimization of AI-based CT quantitative analysis technology in the field of COVID-19 pneumonia diagnosis and treatment. First, the sample size of the study is relatively small, especially the number of critical type patients is only 4 cases, which may lead to a certain bias in the research results, and the statistical power of the study needs to be further improved by expanding the sample size in subsequent research. Second, the study only focused on the quantitative analysis of the volume of total lung infection, GGO and SO lesions, and did not involve the in-depth analysis of the nature, density, morphology and texture information of the lesions. In the future, the AI system needs to be optimized to integrate more imaging features for comprehensive analysis, so as to further improve the accuracy of predicting the clinical classification and disease progression of COVID-19 pneumonia. Third, the study is a single-center retrospective study, and the research results need to be verified by multi-center, prospective clinical studies to improve the external validity and clinical applicability of the results.

Despite these limitations, the study has important academic and clinical value, and its research results have far-reaching implications for the application of AI technology in the medical imaging diagnosis of infectious diseases and the clinical management of COVID-19 pneumonia. First, the study confirmed the reliability and validity of AI-based CT quantitative analysis in the assessment of COVID-19 pneumonia clinical classification, providing a new objective imaging method for the clinical severity assessment of COVID-19 pneumonia, and making up for the deficiencies of traditional subjective CT image analysis. Second, the study provides a standardized technical protocol for AI-based CT quantitative analysis of COVID-19 pneumonia, including CT scanning parameters, post-processing methods, and AI quantitative indicator selection, which is conducive to the popularization and application of this technology in clinical practice. Third, the study lays a foundation for the further development of AI medical imaging technology in the field of COVID-19, and indicates that with the continuous optimization of AI algorithms and diagnostic tools, and the accumulation of large-sample quantitative analysis data of COVID-19, AI technology will play a more important role in the differential diagnosis, precise quantitative analysis, and disease progression prediction of COVID-19 pneumonia CT imaging. Fourth, the research results also provide a reference for the application of AI technology in the diagnosis and treatment of other infectious lung diseases, and open up a new path for the integration of AI and medical imaging in the field of infectious diseases.

In the post-pandemic era, the research and application of AI technology in the field of medical imaging are still in a stage of rapid development. The global spread of COVID-19 has not only accelerated the clinical transformation of AI medical imaging technology, but also put forward higher requirements for its accuracy, intelligence and clinical applicability. For COVID-19 pneumonia, in addition to the assessment of clinical classification, the follow-up research of AI-based CT quantitative analysis can be further expanded to the fields of disease progression prediction, curative effect evaluation, and prognosis judgment. For example, by dynamically monitoring the changes of AI-derived CT quantitative indicators such as GGO volume and SO volume during the treatment of COVID-19 patients, clinicians can evaluate the curative effect of the treatment regimen in real time and adjust the treatment plan in a timely manner; by establishing a prediction model based on AI CT quantitative indicators and clinical factors (such as age, laboratory indicators), the risk of disease progression and poor prognosis of COVID-19 patients can be predicted, so as to implement personalized precise treatment. In addition, the application of low-dose CT in the screening and follow-up of COVID-19 patients, and the research on the permanent lung damage caused by COVID-19 infection are also important directions for the combination of AI and CT imaging technology in the future.

The integration of AI and medical imaging is an important trend in the development of modern medicine, and the study on AI-powered CT quantitative analysis for predicting and evaluating the clinical classification of COVID-19 pneumonia is a typical example of the successful application of this trend in the field of infectious diseases. With the continuous progress of AI technology, the continuous accumulation of medical imaging big data, and the deep integration of clinical medicine and medical engineering, AI-based medical imaging quantitative analysis technology will be more widely used in the diagnosis and treatment of various diseases, bringing new opportunities for the development of precision medicine and intelligent medicine. In the face of the long-term impact of the COVID-19 pandemic and the potential threat of other emerging infectious diseases, the research and application of AI medical imaging technology will continue to be an important research focus in the global medical field, providing a strong technical support for the early diagnosis, severity assessment, and precise treatment of infectious diseases, and making important contributions to the protection of global public health and human life health.

Author Affiliations and Publication Information Authors: Liu Li, Chen Hong, Zhong Wei, Wei Dongmei, Song Yongli, Li Huixin, Zhou Xin, Gao Xiaolong

1. CRT Clinical R&D Group, The First Hospital of Qiqihar, Qiqihar 161005, Heilongjiang, China;
2. Hangzhou Yitu Medical Technology Co., Ltd., Hangzhou 310024, Zhejiang, China;
3. School of Medical Technology, Qiqihar Medical University, Qiqihar 161000, Heilongjiang, China Journal: CT Theory and Applications Volume 30, Issue 6, December 2021, Pages 743-751 DOI: 10.15953/j.1004-4140.2021.30.06.10

(Word count: 3896)