|Year : 2022 | Volume
| Issue : 3 | Page : 86-91
Progress of artificial intelligence in imaging for the diagnosis of drug-resistant pulmonary tuberculosis
Fengli Jiang1, Yu Wang1, Chuanjun Xu2, Qiuzhen Xu1
1 Department of Radiology, Zhongda Hospital, Medical School, Southeast University, Nanjing, China
2 Department of Radiology, The Second Hospital of Nanjing, Nanjing University of Chinese Medicine, Nanjing, China
|Date of Submission||29-Jul-2022|
|Date of Acceptance||10-Sep-2022|
|Date of Web Publication||22-Dec-2022|
Department of Radiology, The Second Hospital of Nanjing, Nanjing University of Chinese Medicine, Nanjing
Department of Radiology, Zhongda Hospital, Medical School, Southeast University, Nanjing
Source of Support: None, Conflict of Interest: None
Recent technical advances have led to the application of artificial intelligence in many areas of medical science. This approach was applied early on to medical imaging, which involves a large amount of data for diagnosis. The application of artificial intelligence and imaging diagnostics for disease screening, diagnosis, and prognosis prediction is an area of active research. Early diagnosis and effective management of drug-resistant pulmonary tuberculosis (TB) can effectively control the spread of Mycobacterium TB, reduce hospitalization, and improve prognosis. We review the progress of artificial intelligence in assisting imaging-based diagnosis of this disease, and we offer useful perspectives on future research in this area.
Keywords: Artificial intelligence, drug resistance, pulmonary tuberculosis
|How to cite this article:|
Jiang F, Wang Y, Xu C, Xu Q. Progress of artificial intelligence in imaging for the diagnosis of drug-resistant pulmonary tuberculosis. Radiol Infect Dis 2022;9:86-91
|How to cite this URL:|
Jiang F, Wang Y, Xu C, Xu Q. Progress of artificial intelligence in imaging for the diagnosis of drug-resistant pulmonary tuberculosis. Radiol Infect Dis [serial online] 2022 [cited 2023 Jun 3];9:86-91. Available from: http://www.ridiseases.org/text.asp?2022/9/3/86/364776
| Introduction|| |
Tuberculosis (TB) is a chronic infection caused by Mycobacterium TB (MTB). Its most common manifestation is pulmonary TB. Starting with 2007, TB has surpassed AIDS in becoming the most deadly infectious disease, and, more generally, a leading cause of death worldwide. On October 14, 2021, the global TB report 2021 released by the World Health Organization estimated 9.9 million cases of TB worldwide in 2020. Globally, the burden of rifampicin-resistant or multidrug-resistant TB (MDR-TB) is stable. For more than 10 years, the best estimate of the proportion of people diagnosed with TB for the first time who had MDR/RR-TB has remained at about 3%–4%. Compared with 2019, the number of patients diagnosed with drug-resistant (DR) TB decreased by 22% from a total of 201,997 cases to 132,222 cases of MDR/RR-TB, and 25,681 instances of preextensively DR TB (pre-XDR-TB) or XDR-TB.
DR-TB is a TB infection that is resistant to one or more anti-TB drugs. Treatment of DR-TB patients requires second-line anti-TB medications, such as fluoroquinolones and bedaquiline. Administration of these drugs involves several procedural difficulties, longer hospitalization times, and a higher risk of serious adverse effects, thereby posing new challenges to successful global TB prevention and control.
The microbiological examination is an important method for the diagnosis of TB. Early diagnosis of DR-TB is essential for selecting an effective treatment regimen. The procedures involved in growing cultures of MTB from clinical samples, and in performing the current “gold standard” of phenotypic drug susceptibility testing (DST), are time-consuming (generally 812 weeks) and often only conducted after the failure of initial conventional treatment. In 2019, ATS/CDC/ERS/IDSA clinical practice guidelines for the treatment of DR-TB emphasized the importance of molecular DST. They proposed that drug resistance should always be considered a possibility, and that all patients should undergo rapid molecular DST. Although molecular testing greatly reduces detection time, it has the disadvantages of high technical requirements, high cost, and poor deployment to remote areas. In addition, resistance genes have not been fully discovered, which may lead to false-negative results in the detection of some DR MTB.
The number of MDR-TB/RR-TB patients detected and registered worldwide between 2018 and 2020 accounted for only 32% of the estimated total. This might be due to the low specificity and sensitivity of native diagnostic tests, as well as their inability to detect TB and treatment resistance in a timely manner. Therefore, there is an urgent need for improving the ability to diagnose DR-TB patients, so as to effectively treat them and stop the spread of MTB.
Chest imaging examinations are essential in the early detection of pulmonary TB and therapeutic drug monitoring. Chest computed tomography (CT) findings,, can be used to distinguish drug-sensitive TB (DS-TB) from DR-TB. When multiple CT signs of DR pulmonary TB coexist, they can serve as starting points for the timely detection of suspected DR-TB patients and can effectively improve clinical diagnosis efficiency, thereby supporting individually tailored treatment plans and improved efficacy. However, the imaging characteristics of DS and DR pulmonary TB overlap considerably, making it difficult to distinguish the two variants based on subjective visual observation. Because artificial intelligence (AI) algorithms can learn autonomously and mine information that cannot be identified by the naked eye in medical images, they represent an important tool for assisting in the diagnosis of DR pulmonary TB. Scientists have conducted preliminary research on AI technology in assisting the differential diagnosis of DR and DS pulmonary TB. Below we review progress in this area.
| Overview of AI Applications in Chest Imaging|| |
AI belongs to the field of computer science, which refers to a machine with human intelligence when performing specific tasks. Machine learning (ML) is a practical approach to the implementation of artificial intelligence (AI), which “trains” the model to learn how to complete tasks by loading a large amount of data into the model and carrying out self-directed learning of the data, without the need for computer algorithms specifically tailored to the problem at hand.
AI algorithms that mine medical imaging information for aiding clinical decisions are increasingly being used in clinical practice. The availability of large diagnostic and medical imaging datasets has brought new opportunities for radiologists to perform AI research. At present, various AI algorithms have been developed and tested on medical images for multiple applications, such as lesion detection, lesion segmentation, lesion characterization, and prediction of prognosis.
AI carries significant potential for auxiliary diagnosis of chest imaging. Moreover, AI can accurately detect lung nodules and breast masses,, assist low-dose CT in the early screening of lung cancer patients, and aid in X-ray screening of breast cancer patients. For instance, Setio et al. found that computer-aided systems for the detection of pulmonary nodules from CT scans reached sensitivity values of 98.3% (234/238) and 94.1% (224/238) at an average of 4.0 and 1.0 false positives per scan, respectively. The above systems can identify the vast majority of highly suspicious lesions in thoracic CT scans with a small number of false positives. By training on massive image datasets, AI can also perform differential diagnosis for different lung diseases. For example, Wang et al. used deep learning to identify patients who may have COVID-19 at fever clinics. Their algorithms produced an area under the curve (AUC) of 0.953, the sensitivity of 0.923, specificity of 0.851, positive predictive value of 0.790, and negative predictive value of 0.948. Feng et al. reported AUC values of 0.809, 0.889, and 0.879 for cohorts used during external validation, training, and internal validation of a CT-based deep learning nomogram for the diagnosis of tuberculous granuloma and lung cancer. Taken together, these findings indicate the accessory diagnostic value of AI. In terms of lesion segmentation, AI algorithms can successfully segment lesions within the lung, allowing accurate calculation of lesion volume and density. A possible diagnosis can be proposed by extracting the high-throughput radiomic features of the lesions, and by evaluating the curative effect and the prognosis of patients.
| AI-assisted Imaging of Drug-resistant Pulmonary Tuberculosis|| |
AI algorithms for screening of pulmonary tuberculosis based on chest radiographs
Chest radiography is a typical imaging technique for screening pulmonary TB, and it plays a significant role in medical practice. However, chest radiographs are two-dimensional images of a three-dimensional structure and therefore provide poor representations of overlapping anatomical elements, making it difficult to diagnose abnormal structures. Moreover, different radiologists may reach different diagnoses because of differences in their professional knowledge, and because of contingent factors such as fatigue and distraction. These factors often lead to the inconsistent and subjective interpretation of chest radiographs. The adoption of software programs for computer-aided diagnosis to automatically read chest radiographs can avoid inter-observer differences and reduce detection errors. Pulmonary TB screening with AI based on chest X-ray (CXR) images has achieved remarkable results.,, Therefore, AI-aided detection of drug-resistant pulmonary TB is expected to alleviate the shortage of radiology technicians in local hospitals. It can also provisionally identify suspected drug-resistant pulmonary TB for further inspection, thus reducing the workload of doctors.
Several studies have reported good diagnostic efficiency for AI-assisted diagnosis of drug-resistant pulmonary TB based on chest radiography. Yang et al. trained machine classifiers on chest radiographs and clinical data from 782 DS-TB and 1455 DR-TB cases using 26 radiological features and three clinical features from the dataset. When combining all 25 statistically relevant features, the average accuracy and AUC values for automatic discrimination between DR-TB and DS-TB were 72.34% and 78.42%, respectively, as returned by ten-fold cross-validation using a support vector machine. However, the performance of their machine classifier to identify DR-TB from other countries declined because it predominantly learned DS and drug-resistant features from six nations. Moreover, the implementation of their ML model cannot be fully automated because their best-performing model requires radiological findings to be evaluated by a radiologist.
When attempting AI-assisted diagnosis of drug-resistant pulmonary TB from chest radiographs (CXRs), deep learning faces significant challenges in the prediction of cavities, hilar lymphadenopathy, pleural effusion, and atelectasis. Engle et al. examined 12 Qure. ai classifiers (deep learning classifiers) using the TB Portals Program (TBPP) database. With the aid of 22 X-ray machine models from six vendors, the Qure. AI classifiers were trained on 1.2 million CXRs. The images were down-sampled and resized to a common standard format because they originally supported different resolutions and quality. Preprocessing also involved several data augmentation techniques specific to the abnormalities present in the CXR images. The results showed the prediction of cavities, nodules, pleural effusions, and hilar lymphadenopathy with Qure. AI classifiers matched the annotations made by human experts.
Jaeger et al. tested different ML architectures for discriminating between DS-TB and DR-TB, including convolutional neural networks, support vector machines, an artificial neural network (ANN) reliant on shape and texture features, and a classifier based on the pretrained VGG-v16 network. The ANN with shape and texture features performed best, with an AUC value of 65% for a dataset consisting of 135 participants. Therefore, information about DR-TB from conventional CXR that cannot be seen by the naked eye can nonetheless be exploited by AI algorithms. Karki et al. further expanded the dataset. These authors used standard and specific convolutional neural networks to classify patients from multiple databases (1821 DR-TB patients and 1821 DS-TB patients), and employed both classical and deep learning-based data augmentation techniques to enhance classification. Their classification performance improved on the state-of-the-art, with values for area under the ROC curve of up to 85% when using a pretrained Inception V3 network.
Numerous papers on the diagnosis of pulmonary TB with AI-aided interpretation of chest radiographs used the same database to train and test models. Karki et al. utilized data from the TB portal dataset (containing 1756 patients from 10 countries) and publicly available TB X-ray data collected from a Chinese hospital (662 patients). They found that the generalization performance of the classifier was poorer (AUC 65%) compared with the cross-validation performance (AUC 79%). These discrepancies may be attributed to large differences in standards and quality for images generated by different medical imaging modalities and different hardware. The ability of AI to support chest radiographs in the diagnosis of pulmonary TB still faces significant challenges.
AI-assisted screening of pulmonary tuberculosis based on computed tomography
There are relatively few AI studies on the use of CT scans to categorize pulmonary TB, but existing reports on AI-assisted differential diagnosis of pulmonary TB and other pulmonary diseases from CT images can offer useful ideas for future classification of pulmonary TB. Ma et al. retrospectively collected a clinical CT imaging dataset consisting of 846 patients. Independent subsets of training and testing data were created from the dataset. The testing dataset contained 139 cases of active TB (ATB), 40 cases of pneumonia, and 100 normal cases, while the training dataset contained 337 cases of ATB, 110 cases of pneumonia, and 120 normal cases. The authors used a U-Net deep learning system for the automatic detection and segmentation of ATB lesions. Because CT scanning produces tomographic images, which differ from X-ray images, the authors carried out training of the deep learning algorithm by marking lesions on the CT image, and by transforming the two-dimensional lesions into three-dimensional lesions via a clustering method based on connectivity within a specified region of interest. The AI algorithm produced an AUC value of 0.980 for a separate test dataset. These results show that AI can adequately diagnose and differentiate pulmonary TB, and provide useful indications for exploiting this technology to distinguish drug-resistant from DS pulmonary TB in the future. Wang et al. collected chest CT scans of 804 TB patients and 301 patients without TB to construct and evaluate a 3D RESNET deep learning framework. They reported AUC values of 0.9, 0.88, and 0.86 for the training set, validation set, and test set, respectively. The AUC value for the external test was 0.78. Without manual labeling, the model automatically recognized anomalous lung regions more effectively and with higher performance than the radiologists. Building on these promising results, efforts are underway to design a deep learning algorithm for recognizing the therapeutic resistance of TB, to simplify the arduous and time-consuming steps associated with this process, and to prevent inaccuracy in labeling TB lesions.
Several different algorithms and models have been used for AI-assisted diagnosis of drug-resistant pulmonary TB from CT images. Each algorithm not only presents some advantages but also presents disadvantages. Cid et al. proposed a general pipeline to automatically obtain a texture-based graph model of the lungs and used SVM classifiers to identify MDR TB based on the CT images of 230 drug-resistant pulmonary TB from the ImageCLEF 2017 competition. Their model obtained 51.64% prediction accuracy for MDR-TB diagnosis. In related research, Lu et al. addressed this classification task using data from the ImageCLEF 2017 competition. They chose a deep learning neural network as classifier, more specifically, a pretrained VGG16 model that had been modified to extract features to address the issue of small sample categorization. Their method outperformed the top competitor with an accuracy rate of 64%. In separate work, Gentili et al. reformatted the CT images to the coronal plane and used a pretrained ResNet50 CNN, while Cid et al. used a network model reliant on 3D texture information combined with support vector machines. They competed in the same challenge, which involves differentiating between DR-TB and DS-TB using thoracic CT images from the ImageCLEF 2018 competition. The selected dataset included 236 test images and 259 training images with approximately 50% DR-TB cases. Unfortunately, the results were unsatisfactory, with AUC values around 0.6. Following these disappointing attempts during the 2017 and 2018 competitions, the organizers subsequently deleted the subtask of classifying drug resistance of pulmonary TB based on CT images, and concluded that the images alone were insufficient to complete the MDR subtask.
To assess the efficacy of treatment using related tools, Gao et al. aimed to distinguish MDR patients from DS patients using CT lung scans. They relied on a collection of datasets from 230 patients from the ImageCLEF 2017 competition to develop a patch-based deep CNN combined with an SVM classifier, and used this approach to examine smaller datasets (i.e., hundreds) containing TB-specific CT features within regions restricted to abnormalities. Their results demonstrated that the suggested integration of CNN with SVM and patch-based analysis outperforms CNN alone, with the best classification accuracy rate reaching 91.11%. Notwithstanding the significant challenges associated with distinguishing between DS-TB and DR-TB using AI-based analysis of CT images, the findings reviewed above are encouraging, and it is expected that they will improve further with the introduction of larger training sets.
Image database of drug-resistant pulmonary tuberculosis
At present, AI-assisted imaging research is mainly based on single-center data with a small sample size, and there is no external or prospective data to verify the effectiveness of AI algorithms. Therefore, at this stage, it is important that different groups share multi-center imaging data pertaining to drug-resistant pulmonary TB. The NIAID TBPP, is a public dataset that was introduced in 2012 with the goal of assembling demographic, geographical, medical, laboratory, and imaging data (including CT scan and X-ray data) from DR-TB patients, alongside relevant pathogen genomic information. As of June 2022, the database has collected 7500 TB cases from 11 countries. Because this database consists primarily of DR-TB cases, the relative representation of DR-TB and DS-TB is unbalanced. The TBPP platform transforms images into data suitable for statistical comparison using professional annotations and computer-generated descriptions. On the data page of the TB Portals Data Browser (https://data.tbportals.niaid.nih.gov), users can perform multiple keyword searches for defined clinical cases. Therefore, TBPP provides a useful platform for data mining and AI algorithm testing.
| Conclusions|| |
DR-TB remains the most challenging problem in TB prevention and control. Imaging data contain extensive information about TB drug resistance. This source of information can be mined using AI to aid clinical decision-making. The available published datasets consist mainly of annotated chest radiographs. As a consequence, AI-assisted diagnosis of drug-resistant pulmonary TB is primarily based on plain CXR. Accurate labeling of relevant lesions is crucial for successful classification by the AI algorithm, which adds to the labor associated with data annotation. However, at present, CT examination is usually recommended for most hospitalized patients. Future studies should focus on AI-guided analysis of CT images to quickly distinguish between DS-TB and DR-TB before treatment, so as to better assist with the subsequent clinical diagnosis and targeted treatment. An additional challenge is represented by the diversity of model specifications associated with different medical imaging equipment, and by the different detection parameters adopted by different hospitals. This lack of standardization makes it difficult to normalize data sampling. In the future, we should therefore aim to standardize and collect large amounts of high-quality medical imaging data, which will serve as training datasets for enhancing the generalization and robustness of AI models.
We would like to thank Liwen Bianji (Edanz) (www. liwenbianji. cn) for editing the English text of a draft of this manuscript.
Financial support and sponsorship
Supported by a grant from Infectious and Inflammatory Radiology Committee of Jiangsu Research Hospital Association (2022-FS-01-001; 2022-FS-01-002)and Supported by Nanjing Medical Science and Technique Development Foundation (YKK21126).
Conflicts of interest
There are no conflicts of interest.
| References|| |
World Health Organization. Global Tuberculosis Report 2021 [M]. Licence: CC BY-NC-SA 3.0 IGO. Geneva: World Health Organization; 2021.
Nahid P, Mase SR, Migliori GB, Sotgiu G, Bothamley GH, Brozek JL, et al.
Treatment of drug-resistant tuberculosis. An official ATS/CDC/ERS/IDSA clinical practice guideline. Am J Respir Crit Care Med 2019;200:e93-142.
Suárez I, Fünger SM, Kröger S, Rademacher J, Fätkenheuer G, Rybniker J. The diagnosis and treatment of tuberculosis. Dtsch Arztebl Int 2019;116:729-35.
Tornheim JA, Starks AM, Rodwell TC, Gardy JL, Walker TM, Cirillo DM, et al.
Building the framework for standardized clinical laboratory reporting of next-generation sequencing data for resistance-associated mutations in Mycobacterium tuberculosis
complex. Clin Infect Dis 2019;69:1631-3.
Van Rie A, Enarson D. XDR tuberculosis: An indicator of public-health negligence. Lancet 2006;368:1554-6.
Wáng YX, Chung MJ, Skrahin A, Rosenthal A, Gabrielian A, Tartakovsky M. Radiological signs associated with pulmonary multi-drug resistant tuberculosis: An analysis of published evidences. Quant Imaging Med Surg 2018;8:161-73.
Li D, He W, Chen B, Lv P. Primary multidrug-resistant tuberculosis versus drug-sensitive tuberculosis in non-HIV-infected patients: Comparisons of CT findings. PLoS One 2017;12:e0176354.
Cheng N, Wu S, Luo X, Xu C, Lou Q, Zhu J, et al.
A comparative study of chest computed tomography findings: 1030 cases of drug-sensitive tuberculosis versus 516 cases of drug-resistant tuberculosis. Infect Drug Resist 2021;14:1115-28.
Pesapane F, Codari M, Sardanelli F. Artificial intelligence in medical imaging: Threat or opportunity? Radiologists again at the forefront of innovation in medicine. Eur Radiol Exp 2018;2:35.
Willemink MJ, Koszek WA, Hardell C, Wu J, Fleischmann D, Harvey H, et al.
Preparing medical imaging data for machine learning. Radiology 2020;295:4-15.
Setzer FC, Shi KJ, Zhang Z, Yan H, Yoon H, Mupparapu M, et al.
Artificial intelligence for the computer-aided detection of periapical lesions in cone-beam computed tomographic images. J Endod 2020;46:987-93.
Rodríguez-Ruiz A, Krupinski E, Mordang JJ, Schilling K, Heywang-Köbrunner SH, Sechopoulos I, et al.
Detection of breast cancer with mammography: Effect of an artificial intelligence support system. Radiology 2019;290:305-14.
Dembrower K, Liu Y, Azizpour H, Eklund M, Smith K, Lindholm P, et al.
Comparison of a deep learning risk score and standard mammographic density score for breast cancer risk prediction. Radiology 2020;294:265-72.
Setio AA, Jacobs C, Gelderblom J, van Ginneken B. Automatic detection of large pulmonary solid nodules in thoracic CT images. Med Phys 2015;42:5642-53.
Wang M, Xia C, Huang L, Xu S, Qin C, Liu J, et al.
Deep learning-based triage and analysis of lesion burden for COVID-19: A retrospective study with external validation. Lancet Digit Health 2020;2:e506-15.
Feng B, Chen X, Chen Y, Lu S, Liu K, Li K, et al.
Solitary solid pulmonary nodules: A CT-based deep learning nomogram helps differentiate tuberculosis granulomas from lung adenocarcinomas. Eur Radiol 2020;30:6497-507.
Yue H, Yu Q, Liu C, Huang Y, Jiang Z, Shao C, et al.
Machine learning-based CT radiomics method for predicting hospital stay in patients with pneumonia associated with SARS-CoV-2 infection: A multicenter study. Ann Transl Med 2020;8:859.
Huang S, Wang Y, Zhou Z, Yu Q, Yu Y, Yang Y, et al.
Distribution atlas of COVID-19 pneumonia on computed tomography: A deep learning based description. Phenomics 2021;1:62-72.
Qin ZZ, Sander MS, Rai B, Titahong CN, Sudrungrot S, Laah SN, et al.
Using artificial intelligence to read chest radiographs for tuberculosis detection: A multi-site evaluation of the diagnostic accuracy of three deep learning systems. Sci Rep 2019;9:15000.
Lakhani P, Sundaram B. Deep learning at chest radiography: Automated classification of pulmonary tuberculosis by using convolutional neural networks. Radiology 2017;284:574-82.
Hwang EJ, Park S, Jin KN, Kim JI, Choi SY, Lee JH, et al.
Development and validation of a deep learning-based automatic detection algorithm for active pulmonary tuberculosis on chest radiographs. Clin Infect Dis 2019;69:739-47.
Yang F, Yu H, Kantipudi K, Karki M, Kassim YM, Rosenthal A, et al.
Differentiating between drug-sensitive and drug-resistant tuberculosis with machine learning for clinical and radiological features. Quant Imaging Med Surg 2022;12:675-87.
Engle E, Gabrielian A, Long A, Hurt DE, Rosenthal A. Performance of Qure.ai automatic classifiers against a large annotated database of patients with diverse forms of tuberculosis. PLoS One 2020;15:e0224445.
Jaeger S, Juarez-Espinosa OH, Candemir S, Poostchi M, Yang F, Kim L, et al.
Detecting drug-resistant tuberculosis in chest radiographs. Int J Comput Assist Radiol Surg 2018;13:1915-25.
Karki M, Kantipudi K, Yu H, Yang F, Kassim YM, Yaniv Z, et al.
Identifying drug-resistant tuberculosis in chest radiographs: Evaluation of CNN architectures and training strategies. Annu Int Conf IEEE Eng Med Biol Soc 2021;2021:2964-7.
Karki M, Kantipudi K, Yang F, Yu H, Wang YX, Yaniv Z, et al.
Generalization challenges in drug-resistant tuberculosis detection from chest X-rays. Diagnostics (Basel) 2022;12:188.
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.
Wang L, Ding W, Mo Y, Shi D, Zhang S, Zhong L, et al.
Distinguishing nontuberculous mycobacteria from Mycobacterium tuberculosis
lung disease from CT images using a deep learning framework. Eur J Nucl Med Mol Imaging 2021;48:4293-306.
Cid YD, Batmanghelich K, Mueller H. Textured Graph-Based Model of the Lungs: Application on Tuberculosis Type Classification and Multi-drug Resistance Detection; Proceedings of the 9th
International Conference of the Conference-and-Labs-of-the-Evaluation-Forum-Association, (CLEF), Avignon, September, 10-14, 2018 [C].FRANCE: Univ Avignon, c2018.Avignon, September, 10-14, 2018 [C].FRANCE:Univ Avignon, c2018.
Lu MC, Gao L, Guo XF, Yin JP. A Framework of Predicting Drug Resistance of Lung Tuberculosis by Utilizing Radiological Images; Proceedings of the 2018 Tenth International Conference on Advanced Computational Intelligence (ICACI), xiamen, March, 29-31, 2018 [C].PEOPLES R CHINA: xiamen,c2018.
Gao XW, Qian Y. Prediction of multidrug-resistant TB from CT pulmonary images based on deep learning techniques. Mol Pharm 2018;15:4326-35.
Gabrielian A, Engle E, Harris M, Wollenberg K, Juarez-Espinosa O, Glogowski A, et al.
TB DEPOT (data exploration portal): A multi-domain tuberculosis data analysis resource. PLoS One 2019;14:e0217410.
Rosenfeld G, Gabrielian A, Wang Q, Gu J, Hurt DE, Long A, et al.
Radiologist observations of computed tomography (CT) images predict treatment outcome in TB Portals, a real-world database of tuberculosis (TB) cases. PLoS One 2021;16:e0247906.
Jiang Y, Zhao X, Fan Z. Intelligence classification algorithm-based drug-resistant pulmonary tuberculosis computed tomography imaging features and influencing factors. Comput Intell Neurosci 2022;2022:3141807.