Artificial intelligence will be a milestone in medical imaging development

Ruili Li; Hongjun Li

doi:10.4103/RID.RID_27_22

REVIEW ARTICLE

Year : 2022 | Volume : 9 | Issue : 3 | Page : 96-99

Artificial intelligence will be a milestone in medical imaging development

Ruili Li¹, Hongjun Li²
¹ Department of Radiology and Nuclear Medicine, Xuanwu Hospital, Capital Medical University; Beijing Key Laboratory of Magnetic Resonance Imaging and Brain Informatics, Beijing, China
² Department of Radiology, Beijing Youan Hospital, Capital Medical University, Beijing, China

Date of Submission	23-May-2022
Date of Acceptance	09-Sep-2022
Date of Web Publication	22-Dec-2022

Correspondence Address:
Hongjun Li
Department of Radiology, Beijing Youan Hospital, Capital Medical University, No. 8 Xi Tou Tiao Youanmen Wai, Fengtai District, Beijing 100069
China

Source of Support: None, Conflict of Interest: None

DOI: 10.4103/RID.RID_27_22

Abstract

Artificial intelligence (AI) is a cutting-edge technology that is applied in many fields. Medical imaging AI is also developing rapidly, and has markedly improved disease detection, diagnosis, prognosis, and monitoring. It also has led to fundamental changes in the way of radiologists' work. The potential new capabilities provided by AI will make the practice of radiology more efficient and effective. Herein, we review the application, current limitations and future opportunities of AI models in medical imaging.

Keywords: Artificial intelligence, medical imaging, radiology

How to cite this article:
Li R, Li H. Artificial intelligence will be a milestone in medical imaging development. Radiol Infect Dis 2022;9:96-9

How to cite this URL:
Li R, Li H. Artificial intelligence will be a milestone in medical imaging development. Radiol Infect Dis [serial online] 2022 [cited 2023 Jun 3];9:96-9. Available from: http://www.ridiseases.org/text.asp?2022/9/3/96/364773

Introduction

With the rapid development of science and technology in the information age, artificial intelligence (AI) has become a hot topic, and plays an important role in many aspects of society.^[1] There is increasing use of AI in the medical field, particularly in medical imaging. Visual evidence-based medical imaging is a critical technology for clinical diagnosis, treatment decision guidance, and efficacy evaluation, and it generates an enormous amount of data in the hospital system. Indeed, medical imaging data have the 5V characteristics of big data, i.e., volume, velocity, variety, value, and veracity-making AI-based medical imaging the first to be developed and applied in various medical services.

AI has enormous potential for use in clinical applications. In China, medical imaging data are growing at an annual rate of 30%, while the annual growth rate of the number of radiologists is only 4%~6%. In addition, traditional radiologists read images by eye to distinguish visual features and make a diagnosis based on experience. Thus, a large number of readings, the experience of the radiologist, and various subjective factors can all contribute to missed diagnoses and affect disease assessment. By contrast, AI-based medical imaging can identify and quantify lesions automatically, which is more accurate, more objective, repeatable, and can reduce missed diagnoses and errors caused by subjective visual assessment, thus improving efficiency and diagnosis accuracy. In turn, this allows radiologists to focus more on the diagnosis.

Brief Overview of Artificial Intelligence

Machine learning (ML) refers to training models that learn features and associated parameters that most likely describe observed data. Classic ML techniques include decision trees, random forests, support vector machines, and dimensionality reduction techniques such as principal component analysis. The computational requirements of these techniques are relatively small because these models involve relatively few parameters compared to the deep learning (DL) algorithm. Radiomics is a classic ML model in AI. In the past, radiomics was used to detect hidden features (e.g., pattern, density, and texture) in images and select the most useful and relevant parameters as biological markers. In recent years, radiomics has been used to (1) provide high-throughput quantitative information about diseases from massive datasets, including radiographic images, digital pathology images and genes, associated radiographic images with genes, and clinical data (e.g., classification, efficacy, and prognosis), (2) integrate multiple models to improve image-based diagnosis, efficacy assessment, and prognosis/prediction, (3) extend its application from morphology to genetics, and (4) aid in personalized precision treatment.^[2],[3]

DL is a ML technology based on a deep neural network that mimics but greatly simplifies human neurons in the brain. DL is based on a deep architecture that supports hierarchical learning and progressively extracts features from data from simple to complex abstractions.^[4] DL is a smarter AI technology based on ML that does not require manual definition in advance. DL can automatically learn from data and summarize representative features. This data-driven model has a compatible effect on abstract features, resulting in more comprehensive generality. It has also been reported that DL achieves a similar diagnostic capacity to imaging specialists in lesion identification, imaging feature analysis, and lesion staging^[5],[6] and can significantly improve the efficiency of imaging diagnosis.

Supervised learning refers to the training program in which the training model must observe the training data and the corresponding ground truth of the data (or sometimes referred to as the “target”).^[7] For example, in mammography, the cancer is accurately labeled on the image (as a label), so the algorithm can “learn” the features of malignant tumors from the labeled annotations. Compared to unsupervised learning, semi-supervised learning, or weakly supervised learning, supervised learning seems to be the most popular method in image classification tasks.

Convolutional Neural Networks (CNNs) consist of a special ML structure, which is one of the most popular structures in current medical image analysis applications.^[4] Training CNNs requires a large amount of data, i.e., much more data than other types of ML. CNNs typically perform end-to-end supervised learning on labeled data, while other architectures perform unsupervised learning tasks on unlabeled, unlabeled data. CNNs consist of a number of layers, of which the “hidden layer” can perform feature extraction and aggregation through convolution and pooling operations. The fully linked layers can perform advanced inference before outputting the final results. Some studies have shown that DL methods have excellent performance in segmentation tasks in magnetic resonance imaging (MRI)^[8] and staging tasks in computed tomography.^[9] The integration of ML, DL, radiomic and CNNs methods has shown a direct relationship between tissue image phenotyping and histobiology, which has important clinical significance.

Application of Artificial Intelligence in Medical Imaging

In medical imaging, physicians usually detect, characterize, and monitor diseases through visual evaluation of medical images. Sometimes, such visual assessments based on professional knowledge and experience may be personal and inaccurate. AI can conduct the quantitative evaluation by automatically identifying imaging information instead of such qualitative reasoning. Therefore, AI can help physicians make more accurate imaging diagnosis, and also reduce the workload of physicians. The AI model has been widely trialled clinically with positive outcomes. A range of AI applications have been used in medical imaging of various human body systems, including in clinical diseases. For example, AI applications in thoracic imaging include lung nodule screening, lung nodule volumetry, detection of pneumonia, tuberculosis diagnosis, detection of common chest radiograph anomalies, emphysema analysis, COVID-19 analysis, pulmonary fibrosis analysis, staging of COPD, prediction of acute respiratory distress and mortality in smokers, automated classification of fibrotic lung diseases, detection and quantification of infiltrative lung diseases, breast nodule screening, and identification of benign and malignant lesions.^{[10],[11],[12]} The cardiovascular imaging applications mainly include image processing, detection, segmentation, coronary artery calcium, coronary plaque detection, plaque properties, plaque phenotyping, coronary artery stenosis, hemodynamic assessment of coronary lesions, myocardial perfusion, myocardial tissue characterization, cardiac function, myocardial activity, adipose tissue characterization, radiomic phenotyping of perivascular fat, and cardiac risk prediction.^[13],[14] The abdominal and pelvic imaging applications mainly include liver image quality evaluation, focal liver lesion detection, focal liver lesion evaluation, diffuse liver disease staging, segmentation of the liver or liver vasculature, treatment response prediction, early diagnosis of hepatocellular carcinoma, rectal cancer, and prostate cancer, non-invasive staging, microvascular invasion prediction, prognosis and prediction, and lymph node metastasis diagnosis.^{[15],[16],[17]} The central nervous system applications mainly focus on image analysis, MRI segmentation, functional-area detailing and positioning, gray matter atrophy, and quantitative analysis of white matter injury.^[18],[19] Finally, AI is used in the musculoskeletal system to perform qualitative judgments on fractures, degenerative spinal diseases, rib deformities, joint deformities, skeletal age prediction, and bone tumors.^[20] However, despite this development and application of AI technology in the medical field, it remains too early to measure its overall performance. Thus, further investment is needed to optimize and establish a composite model of factors related to human physiology and pathology based on clinical needs to improve the quality of AI-assisted diagnosis and treatment.

AI technology has changed the mode and efficiency of imaging diagnosis. AI has been used in all aspects of the imaging workflow, including image scanning, rapid high-quality reconstruction, lesion detection, segmentation, quantitative analysis, risk prediction, prognostic judgment, and evaluation of treatment effects. In addition to providing more accurate auxiliary information for medical decision-making, AI can integrate with the 5G network to promote the implementation and improvement of hierarchical diagnosis and treatment systems via cloud consulting services, increase the availability of high-quality medical resources, and reduce the imbalance between supply and demand of medical resources in China. AI can also reduce the admission pressure in large hospitals, as well as improve the medical level of small hospitals and the overall efficiency of medical treatment.^[21] Furthermore, AI plays an important role in solving the problem of imbalanced regional medical resources.

It is important to note that AI technology is still in the rapid development phase, and the translation from theoretical models to clinical practice remains challenging. Furthermore, the clinical application of AI in medical imaging remains in its infancy, with many technical problems and challenges to be solved,^[22],[23] which include: (1) a lack of high-quality, standardized image training datasets, and of uniform industry standards and management norms^;[24] (2) because of the low contrast of medical images, the boundary between normal and abnormal tissues is blurred, while the applicability of typical DL methods for fine structures such as nerves and blood vessels needs improvement; (3) most AI software for medical imaging is limited to a single model and lacks a DL method that integrates multiple models, while for radiologists, there is a diverse and complicated range of diseases assessed by the same examinations;^[25] and (4) the AI operation process and its logic are not yet well known (termed the “black box”^[26]), while for patient safety, transparency of an AI system operation is an important issue. These difficulties highlight the directions required for future studies.

As part of the future of AI technology, modeling will continue to develop towards product personalization, diversification, multi-modality, multi-model integration, multitasking, integration, and a comprehensive full-process platform. This requires joint efforts in industry, education, research, and clinical studies, with a broad range of applications expected.

Conclusion

Although AI in medical imaging offers many advantages, the current low level of scientific and technological development is an obstacle to the development of comprehensive and high-quality clinical AI applications. Therefore, further studies are needed to improve our knowledge and develop innovative AI technologies for radiological use. AI also provides an opportunity to improve clinical skills, and radiologists who master AI technology will be able to adapt to the evolving field of intelligent medical care. Whether AI will replace radiologists remains unclear, although radiologists who use AI technology in the future will definitely replace those who do not. Proper use of AI technology will improve work efficiency, increase diagnostic accuracy, and help in preclinical or early diagnosis of patients who are asymptomatic and without signs, which outpatient physicians are unable to do. AI can also detect diseases that are not visible to the naked eye and improve the ability to assess disease prognosis and treatment. Thus, innovative research in AI in the medical field will ultimately benefit patients through the development of personalized precision diagnosis and treatment. The national and international collaboration between radiologists, informaticians, academia, and industry is necessary to ensure that AI can be applied more quickly, safely, and effectively in clinical practice.

Financial support and sponsorship

This work was supported by the National Natural Science Foundation of China (grant no. 82202118, 82271963, 61936013), Beijing Natural Science Foundation (7212051), and the Capital Medical University research and incubation fund (grant no. PYZ21129).

Conflicts of interest

There are no conflicts of interest.

References

1.	Russell S, Bohannon J. Artificial intelligence. Fears of an AI pioneer. Science 2015;349:252.
2.	Vaidya P, Bera K, Gupta A, Wang X, Corredor G, Fu P, et al. CT derived radiomic score for predicting the added benefit of adjuvant chemotherapy following surgery in stage I, II resectable non-small cell lung cancer: A retrospective multicohort study for outcome prediction. Lancet Digit Health 2020;2:e116-28.
3.	Kocak B, Durmaz ES, Ates E, Sel I, Turgut Gunes S, Kaya OK, et al. Radiogenomics of lower-grade gliomas: Machine learning-based MRI texture analysis for predicting 1p/19q codeletion status. Eur Radiol 2020;30:877-86.
4.	LeCun Y, Bengio Y, Hinton G. Deep learning. Nature 2015;521:436-44.
5.	Liu X, Faes L, Kale AU, Wagner SK, Fu DJ, Bruynseels A, et al. A comparison of deep learning performance against health-care professionals in detecting diseases from medical imaging: A systematic review and meta-analysis. Lancet Digit Health 2019;1:e271-97.
6.	van Kempen EJ, Post M, Mannil M, Witkam RL, Ter Laan M, Patel A, et al. Performance of machine learning algorithms for glioma segmentation of brain MRI: A systematic literature review and meta-analysis. Eur Radiol 2021;31:9638-53.
7.	Wang S, Summers RM. Machine learning and radiology. Med Image Anal 2012;16:933-51.
8.	Ghafoorian M, Karssemeijer N, Heskes T, van Uden IW, Sanchez CI, Litjens G, et al. Location sensitive deep convolutional neural networks for segmentation of white matter hyperintensities. Sci Rep 2017;7:5110.
9.	González G, Ash SY, Vegas-Sánchez-Ferrero G, Onieva Onieva J, Rahaghi FN, Ross JC, et al. Disease staging and prognosis in smokers using deep learning in chest computed tomography. Am J Respir Crit Care Med 2018;197:193-203.
10.	Liu Y, Wang H, Li Q, McGettigan MJ, Balagurunathan Y, Garcia AL, et al. Radiologic features of small pulmonary nodules and lung cancer risk in the national lung screening trial: A nested case-control study. Radiology 2018;286:298-306.
11.	Zhou J, Zhang Y, Chang KT, Lee KE, Wang O, Li J, et al. Diagnosis of benign and malignant breast lesions on DCE-MRI by using radiomics and deep learning with consideration of peritumor tissue. J Magn Reson Imaging 2020;51:798-809.
12.	Li H, Mendel KR, Lan L, Sheth D, Giger ML. Digital mammography in breast cancer: Additive value of radiomics of breast parenchyma. Radiology 2019;291:15-20.
13.	Narang A, Bae R, Hong H, Thomas Y, Surette S, Cadieu C, et al. Utility of a deep-learning algorithm to guide novices to acquire echocardiograms for limited diagnostic use. JAMA Cardiol 2021;6:624-32.
14.	Augusto JB, Davies RH, Bhuva AN, Knott KD, Seraphim A, Alfarih M, et al. Diagnosis and risk stratification in hypertrophic cardiomyopathy using machine learning wall thickness measurement: A comparison with human test-retest performance. Lancet Digit Health 2021;3:e20-8.
15.	Ioannou GN, Tang W, Beste LA, Tincopa MA, Su GL, Van T, et al. Assessment of a deep learning model to predict hepatocellular carcinoma in patients with hepatitis C cirrhosis. JAMA Netw Open 2020;3:e2015626.
16.	Zhang XY, Wang L, Zhu HT, Li ZW, Ye M, Li XT, et al. Predicting rectal cancer response to neoadjuvant chemoradiotherapy using deep learning of diffusion kurtosis MRI. Radiology 2020;296:56-64.
17.	Mata LA, Retamero JA, Gupta RT, García Figueras R, Luna A. Artificial intelligence-assisted prostate cancer diagnosis: Radiologic-pathologic correlation. Radiographics 2021;41:1676-97.
18.	Xiao B, He N, Wang Q, Cheng Z, Jiao Y, Haacke EM, et al. Quantitative susceptibility mapping based hybrid feature extraction for diagnosis of Parkinson's disease. Neuroimage Clin 2019;24:102070.
19.	Monteiro M, Newcombe VF, Mathieu F, Adatia K, Kamnitsas K, Ferrante E, et al. Multiclass semantic segmentation and quantification of traumatic brain injury lesions on head CT using deep learning: An algorithm development and multicentre validation study. Lancet Digit Health 2020;2:e314-22.
20.	Ahn KS, Bae B, Jang WY, Lee JH, Oh S, Kim BH, et al. Assessment of rapidly advancing bone age during puberty on elbow radiographs using a deep neural network model. Eur Radiol 2021;31:8947-55.
21.	Hosny A, Parmar C, Quackenbush J, Schwartz LH, Aerts HJ. Artificial intelligence in radiology. Nat Rev Cancer 2018;18:500-10.
22.	Harrington SG, Johnson MK. The FDA and artificial intelligence in radiology: Defining new boundaries. J Am Coll Radiol 2019;16:743-4.
23.	Yu KH, Kohane IS. Framing the challenges of artificial intelligence in medicine. BMJ Qual Saf 2019;28:238-41.
24.	Shaw J, Rudzicz F, Jamieson T, Goldfarb A. Artificial intelligence and the implementation challenge. J Med Internet Res 2019;21:e13659.
25.	Heuvelmans MA, Walter JE, Peters RB, Bock GH, Yousaf-Khan U, Aalst CM, et al. Relationship between nodule count and lung cancer probability in baseline CT lung cancer screening: The NELSON study. Lung Cancer 2017;113:45-50.
26.	Ghassemi M, Oakden-Rayner L, Beam AL. The false hope of current approaches to explainable artificial intelligence in health care. Lancet Digit Health 2021;3:e745-50.