|Year : 2022 | Volume
| Issue : 3 | Page : 92-95
Research progress of artificial intelligence-based imaging diagnosis of infectious diseases
Lin Guo1, Li Xia1, Fleming Lure1, Hongjun Li2
1 Shenzhen Zhiying Medical Imaging, Shenzhen, Guangdong, China
2 Department of Radiology, Beijing YouAn Hospital, Capital Medical University, Beijing, China
|Date of Submission||05-Jul-2022|
|Date of Acceptance||31-Aug-2022|
|Date of Web Publication||22-Dec-2022|
Department of Radiology, Beijing You'an Hospital, Capital Medical University, Beijing 100069
Source of Support: None, Conflict of Interest: None
With the rapid development and progress of theory and technology, artificial intelligence (AI) has overcome many early technical limitations. Remarkable advances have recently been made in the application of AI to various areas of health care, and improvements in the performance of computer-aided diagnostics, such as accuracy, specificity, and processing speed, have been achieved, especially in the classification and identification of lesions. We review the various applications and worldwide progress of AI-based imaging diagnosis of infectious diseases.
Keywords: Artificial intelligence, infectious disease, medical imaging
|How to cite this article:|
Guo L, Xia L, Lure F, Li H. Research progress of artificial intelligence-based imaging diagnosis of infectious diseases. Radiol Infect Dis 2022;9:92-5
|How to cite this URL:|
Guo L, Xia L, Lure F, Li H. Research progress of artificial intelligence-based imaging diagnosis of infectious diseases. Radiol Infect Dis [serial online] 2022 [cited 2023 Mar 24];9:92-5. Available from: http://www.ridiseases.org/text.asp?2022/9/3/92/364774
| Introduction|| |
Artificial intelligence (AI) has enabled quantitative interpretations of medical images, which has allowed the extraction of useful information from images to improve diagnostic accuracy. In the field of infectious diseases, AI-based imaging diagnosis studies have achieved promising results, and pulmonary tuberculosis (TB), human immunodeficiency virus (HIV)-combined TB, and coronavirus-19 (COVID-19) are particular research hotspots. The development of AI has involved the integration of multicenter, multimodal, and multi-disease data analyses, and studies have begun focusing on various AI application scenarios, such as screening, lesion grading, and analyzing disease progression and prognosis. Therefore, AI-based imaging studies vary depending on the disease studied. To date, several AI techniques for specific diseases have resulted in the development of commercial products that have obtained certification as Class II and III medical devices. Over the past 2 years, efforts to combat the COVID-19 pandemic have directly promoted the industrialization of these AI products.
| Applications|| |
Application of artificial intelligence in tuberculosis diagnosis
Early screening and diagnosis of TB play an important role in the management and treatment of TB. Clinically, both active pulmonary and drug-resistant TB is generally considered to have a greater risk of spreading; thus, numerous studies have focused on the early screening and auxiliary diagnosis of TB. Approximately 90% of TB originates from high TB burden countries or regions with limited public health provisions. Moreover, these regions often lack experienced radiologists. Therefore, the application of AI can help improve the efficiency of TB management and treatment. For active pulmonary TB, AI systems have been applied to chest X-ray images in the clinic to detect TB. Comparisons between TB detection scores generated by the AI system and those of senior radiologists showed that the AI system was able to detect TB-diagnosed patients more accurately than the senior radiologists. AI has also achieved outstanding detection performance for detecting TB from computed tomography (CT) scans. Ma et al. developed an AI system for TB detection using CT images, and the area under the curve (AUC) of the final model achieved 0.980 on independent testing, with accuracy, sensitivity, and specificity of 0.968, 0.971, and 0.971, respectively. The study demonstrated that AI can be used to distinguish between patients with active and inactive TB, and it is anticipated that AI will help radiologists screen high-risk TB populations and reduce infection and mortality rates.
According to the 2021 Global Tuberculosis Report, the number of drug-resistant TB patients who received treatment in 2020 dropped by 15% since 2019, which equates to approximately one-third of drug-resistant TB patients not receiving effective treatment. Drug resistance testing requires bacteriological confirmation of TB using rapid molecular, culture, and genetic sequencing techniques, which are often costly and time-consuming and are thus not suitable for the timely detection of drug-resistant TB and the adjustment of treatments in advance. However, various techniques, such as texture map models, superpixel methods, and deep learning to identify drug-resistant TB, have been developed. During the 2017 ImageCLEF TB competition, the highest accuracy for the classification of drug-resistant and drug-susceptible TB was 51.64%. To monitor the effectiveness of treatment, AI was used to predict multidrug-resistant patients from CT images, and results showed that the AI model that combined a support vector machine classifier and a deep convolutional neural network could achieve a classification accuracy of 91.11% when applied to the 2017 ImageCLEF competition dataset. It is generally accepted that the identification of patients with drug-resistant TB is challenging for radiologists, and AI may offer novel ideas and approaches for the prediction of drug resistance.
Because of the lack of experienced radiologists, AI has emerged as a valuable potential tool to assist radiologists. However, the World Health Organization emphasizes the importance of conducting independent external tests in target populations before tools are applied in the clinic to ensure that AI systems are both safe and effective. Zhou et al. developed a deep-learning TB screening model using chest radiograph datasets from five centers and independently validated the model using an external dataset comprising over 300,000 radiographs. Results showed that the AI system performed well in the detection of TB, non-TB abnormalities, and healthy chest radiographs. The AUC was >0.9 for the internal dataset and >0.8 for the external dataset. Thus, AI systems have considerable potential as a TB screening tool to provide guidance for further examinations.
Application of artificial intelligence to the diagnosis of human immunodeficiency virus-combined tuberculosis
Comparisons of imaging features between HIV-combined TB and TB have revealed inflammatory exudative changes in the lower lung field, miliary pulmonary TB and hilar, mediastinal lymph node enlargement or extrapulmonary TB, and other atypical signs of TB in patients with HIV-combined TB. However, proliferative and caseous lesions in the upper lung field or lung apex are rare; moreover, features of secondary pulmonary TB, such as cavitation or fibrosis, are also relatively rare. These signs are significantly different from the typical signs of simple pulmonary TB.
Clinically, digital radiography (DR) or CT is typically used to observe the imaging features of different types of TB/HIV-infected patients. However, current imaging methods are not effective because the imaging characteristics of HIV-combined TB are diverse. AI techniques may be valuable auxiliary diagnostic tools for patients with severe TB/HIV infection who require immediate treatment. Furthermore, they may be particularly beneficial for countries where HIV and TB are endemic because of the lack of experienced radiologists who can interpret images, such as chest X-rays. Several studies on the application of AI techniques to the diagnosis of HIV-combined TB have been conducted. In 2021, Wang et al. applied AI to HIV populations for active TB screening and compared the performance between manual and automatic reading by reference to the etiological diagnosis of TB. Results showed that the sensitivity, specificity, and accuracy of manual reading were 70.2%, 71.8%, and 71.7%, respectively, whereas for automatic reading using AI, the sensitivity, specificity, and accuracy were 89.4%, 33.1%, and 37.3%, respectively. The sensitivity of automatic reading using AI was higher than that of manual reading. Another study using deep learning to diagnose TB using chest radiographs of TB/HIV-infected patients showed that the use of deep learning algorithms improved the diagnostic accuracy of radiologists from 0.60 (95% confidence interval [CI] 0.57–0.63) to 0.65 (95% CI 0.60–0.70), which would be beneficial for areas with a high risk of TB/HIV co-infection.
The application of artificial intelligence to the diagnosis of COVID-19
At the beginning of the COVID-19 epidemic, the supply of nucleic acid detection methods barely met demand, and thus CT played an important role in the early screening of suspected cases owing to its high efficiency and sensitivity. Subsequently, molecular detection capabilities improved considerably, and the use of CT imaging gradually shifted toward lesion grading and disease progression and prognosis assessments. For the application of AI to the early screening of COVID-19, Xie et al. used Unet to segment lung images and classify the type of lesions. A comparison of the reading performance of senior radiologists with and without the help of the AI system showed that the two radiologists had a higher performance with the aid of AI, with accuracy increasing from 56.4% and 64.1% to 71.8% and 76.9%, respectively. Similarly, Yang et al. compared the reading performance of radiologists with and without the assistance of an AI system to evaluate the clinical use of AI for COVID-19 diagnosis, and results showed that compared with reading alone, the accuracy and sensitivity of the two radiologists using the AI system increased from 94.1% and 89.5% to 95.1% and 94.2%, respectively. In addition, radiomics, a machine-learning application for medical images, has been used to develop early screening models for COVID-19. Syed et al. trained an AI model based on the CT scans of 1252 patients with COVID-19 and 1229 patients without COVID-19 and obtained an accuracy of 93.7% within 34 s. Fang et al. extracted 23 features from CT images that were highly correlated with COVID-19 symptoms and obtained an AUC of over 0.8 for both the training and test datasets.
In terms of lesion grading and disease progression assessment, Wang et al. reported that conventional imaging staging cannot quantitatively analyze lesion changes to enable clinical classifications. It has been found that patients who present with early symptoms of COVID-19 in imaging may be clinically severe cases, and patients who present with severely progressing symptoms of COVID-19 in imaging may be clinically moderate cases. In addition, conventional imaging staging can be confounded by interobserver variability in judgment. In contrast, AI systems can track pulmonary lesion changes over time and allow more accurate classification between healthy and severe/critical patients. Furthermore, AI techniques can provide quantitative assessments of disease progression. Fuhrman et al. demonstrated that their deep-learning model could identify patients who received and did not receive steroid treatment with an AUC curve of 0.85. This would be valuable for clinical decision-making and evaluating treatment efficiency. To identify predictors for the progression of COVID-19, Li et al. applied AI to CT scans to evaluate lesion involvement, and results showed that consolidation volume was the strongest predictor for disease progression.
For prognostic analysis, Liang et al. developed a prediction model for severe COVID-19 patients based on clinical data that could accurately determine prognosis by calculating the probability that a COVID-19 patient would transition into a critical case. Wang et al. developed an AI prediction model for disease deterioration using CT images and clinical data. From a total of 1051 cases of positive cases from multiple centers, 737 patients were used as the training set, 105 patients were used as the validation set, and 209 patients were used for model testing. Results showed that the AI prediction model successfully classified patients into high-risk and low-risk groups and obtained an AUC of 0.856, an accuracy of 0.833, a sensitivity of 0.622, and a specificity of 0.890. In addition, a radiomics study to predict poor prognoses of COVID-19 patients revealed that radiomics characteristics of chest CTs can accurately predict adverse outcomes in advanced COVID-19 patients. Furthermore, in early-stage COVID-19 patients, radiomics characteristics combined with clinical risk factors could predict adverse prognostic outcomes, which enabled appropriate management and monitoring of COVID-19 patients. An increasing number of AI studies focusing on COVID-19 diagnosis using CT images are being conducted with the aim of improving the efficiency and accuracy of early diagnosis. However, studies investigating treatment assessments, prognosis, and follow-up management remain scarce.
| Research Progress Worldwide|| |
Currently, research in the field of AI-assisted infectious disease diagnosis is primarily focused on a single disease. One reason may be the limited availability of high-quality public datasets to develop AI models. Before the outbreak of COVID-19, TB was the most extensively studied infectious disease owing to the availability of several international public datasets of TB patients that enabled the development and verification of AI models. However, these datasets are not without problems. For example, the quality of labels varies between institutions, and the diagnoses of many cases have not been confirmed pathologically. Moreover, most datasets include only DR images, which limits the development of a multimodal AI technique. However, to meet clinical needs, AI tools have been developed to enable differential diagnoses of various diseases but not to detect a single disease specifically. Numerous studies have yielded promising results in clinical settings; therefore, AI has great potential to be applied en masse in the future. At the beginning of the outbreak of COVID-19, chest imaging became an important tool for diagnosis and treatment evaluation, and imaging diagnostics for the high abundance of COVID-19 cases has further facilitated the development of AI in screening, lesion grading, progression evaluation, prognosis assessment, and follow-up management. This, in turn, has accelerated the development of clinical applications of AI in the field of the infection of COVID-19.
In the future, AI products need to be improved in the following aspects: (1) dataset construction and quality control, which may involve generating high-quality training datasets that comprise information on the number of cases, lesion regions, equipment, scanning parameters, population characteristics, and image staging; (2) data labeling rules, which are a crucial element of AI algorithm development. At present, there is no consensus on data labeling approaches, and thus training of labelers will also be required; (3) comprehensive modeling of imaging, clinical and laboratory examinations, which includes improving physicians' work efficiency, avoiding missed diagnoses, and providing accurate classification and treatment evaluations; and (4) developing more complex algorithms. Because medical data are multi-dimensional, processing and modeling multimodal and multi-task dynamic medical data will require more complex algorithms.
This work was supported by the National Key Research and Development Program of China (grant no. 2019YFE0121400), the Shenzhen Science and Technology Program (grant no. KQTD2017033110081833, JSGG20201102162802008 and JCYJ20220531093817040), and the Shenzhen Fundamental Research Program (grant no. JCYJ20190813153413160). We thank Sarina Iwabuchi, Ph.D., from Liwen Bianji (Edanz) (www.liwenbianji.cn), for editing the English text of a draft of this manuscript.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| References|| |
World Health Organization. WHO Consolidated Guidelines on Tuberculosis: Module 2: Screening: Systematic Screening for Tuberculosis Disease [M]. Geneva: World Health Organization; 2021.
Liao Q, Feng H, Li Y, Lai X, Pan J, Zhou F, et al.
Evaluation of an Artificial Intelligence (AI) system to detect tuberculosis on chest X-ray at a pilot active screening project in Guangdong, China in 2019. J Xray Sci Technol 2022;30:221-30.
Ma L, Wang Y, Guo L, Zhang Y, Wang P, Pei X, et al
. Developing and verifying automatic detection of active pulmonary tuberculosis from multi-slice spiral CT images based on deep learning. J Xray Sci Technol 2020;28:939-51.
Cid YD, Kalinovsky A, Liauchuk V, Kovalev V, Müller H. Overview of ImageCLEF 2017 Tuberculosis Task – Predicting Tuberculosis Type and Drug Resistances [M]. Dublin, Ireland: CLEF 2017 Working Notes; 2017.
Gao XW, Qian Y. Prediction of multidrug-resistant TB from CT pulmonary images based on deep learning techniques. Mol Pharm 2018;15:4326-35.
Zhou W, Cheng G, Zhang Z, Zhu L, Jaeger S, Lure FY, et al
. Deep learning-based pulmonary tuberculosis automated detection on chest radiography: Large-scale independent testing. Quant Imaging Med Surg 2022;12:2344-55.
Lu PX, Yu WY, Zhu WK, Liu Y, Liu JQ. Radiological features of AIDS complicated by pulmonary tuberculosis and the association with CD4+ T lymphocytes. Zhonghua Jie He He Hu Xi Za Zhi 2005;28:13-6.
Wang Q, He J, Li Y, Chen M, Zhou L. A study on the effect of artificial intelligence automatic film reading technology in active tuberculosis screening of HIV/AIDS population. Chin J Antituberc 2021;43:557-61.
Rajpurkar P, O'Connell C, Schechter A, Asnani N, Li J, Kiani A, et al.
CheXaid: Deep learning assistance for physician diagnosis of tuberculosis using chest x-rays in patients with HIV. NPJ Digit Med 2020;3:115.
Xie H, Li Q, Hu PF, Zhu SH, Zhang JF, Zhou HD, et al
. Helping roles of Artificial Intelligence (AI) in the screening and evaluation of COVID-19 based on the CT images. J Inflamm Res 2021;14:1165-72.
Yang Y, Lure FY, Miao H, Zhang Z, Jaeger S, Liu J, et al.
Using artificial intelligence to assist radiologists in distinguishing COVID-19 from other pulmonary infections. J Xray Sci Technol 2021;29:1-17.
Syed HH, Khan MA, Tariq U, Armghan A, Alenezi F, Khan JA, et al.
A Rapid artificial intelligence-based computer-aided diagnosis system for COVID-19 classification from CT images. Behav Neurol 2021;2021:2560388.
Fang M, He B, Li L, Dong D, Yang X, Li C, et al
. CT radiomics can help screen the coronavirus disease 2019 (COVID-19): A preliminary study. Sci China Inf Sci 2020;63:172103.
Yun Y, Wang L, An C, Zhang D, Yang G, Feng K, et al
. Application of artificial intelligence in clinical classification of COVID-19. J Pract Radiol 2021;37:207-10.
Fuhrman JD, Chen J, Dong Z, Lure FY, Luo Z, Giger ML. Cascaded deep transfer learning on thoracic CT in COVID-19 patients treated with steroids. J Med Imaging (Bellingham) 2021;8:014501.
Li Y, Shang K, Bian W, He L, Fan Y, Ren T, et al
. Prediction of disease progression in patients with COVID-19 by artificial intelligence assisted lesion quantification. Sci Rep 2020;10:22083.
Liang W, Yao J, Chen A, Lv Q, Zanin M, Liu J, et al.
Addendum: Early triage of critically ill COVID-19 patients using deep learning. Nat Commun 2021;12:826.
Wang R, Jiao Z, Yang L, Choi JW, Xiong Z, Halsey K, et al
. Artificial intelligence for prediction of COVID-19 progression using CT imaging and clinical data. Eur Radiol 2022;32:205-12.
Wu Q, Wang S, Li L, Wu Q, Qian W, Hu Y, et al
. Radiomics analysis of computed tomography helps predict poor prognostic outcome in COVID-19. Theranostics 2020;10:7231-44.