Abstract:
Author Keywords
COVID-19, artificial intelligence, image acquisition, segmentation, diagnosis
IEEE Keywords
Computed tomography, Artificial intelligence, Image segmentation, Biomedical imaging, X-ray imaging, Three-dimensional displays
Introduction
THE coronavirus disease 2019 (COVID-19), caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), is an ongoing pandemic. The number of people infected by the virus is increasing rapidly. Up to April 9, 2020, 1,436,198 cases of COVID-19 have been reported in over 200 countries and territories, resulting in approximately 85,521 deaths (with a fatal rate of 5.95%) [1]. This has led to great public health concern in the international community, as the World Health Organization (WHO) declared the outbreak to be a Public Health Emergency of International Concern (PHEIC) on January 30, 2020, and recognized it as a pandemic on March 11, 2020 [2, 3].
Reverse Transcription-Polymerase Chain Reaction (RT-PCR) test serves as the gold standard of confirming COVID-19 patients [4]. However, the RT-PCR assay tends to be inadequate in many areas that have been severely hit especially during the early outbreak of this disease. The lab test also suffers from insufficient sensitivity, such as 71% reported in Fang et al. [5]. This is due to many factors, such as sample preparation and quality control [6]. In clinical practice, easily accessible imaging equipment, such as chest X-ray and thoracic CT, provide huge assistance to clinicians [7-12]. Particularly in China, many cases were identified as suspected of COVID-19, if characteristic manifestations in CT scans were observed [6]. The suspected patients, even without clinical symptoms (e.g., fever and coughing), were also hospitalized or quarantined for further lab tests. Given the current sensitivity of the nucleic acid tests, many suspected patients have to be tested multiple times several days apart before reaching a confident diagnosis. Hence, the imaging findings play a critical role in constraining the viral transmission and also fighting against COVID-19.
The workflow of imaging-based diagnosis for COVID-19, taking thoracic CT as an example, includes three stages in general, i.e., 1) pre-scan preparation, 2) image acquisition, and 3) disease diagnosis. In the pre-scan preparation stage, each subject is instructed and assisted by a technician to pose on the patient bed according to a given protocol. In the image acquisition stage, CT images are acquired during a single breath-hold. The scan ranges from the apex to the lung base. Scans are done from the level of the upper thoracic inlet to the inferior level of the costophrenic angle with the optimized parameters set by the radiologist(s), based on the patient’s body shape. From the acquired raw data, CT images are reconstructed and then transmitted through picture archiving and communication systems (PACS) for subsequent reading and diagnosis.
Artificial intelligence (AI), an emerging technology in the field of medical imaging, has contributed actively to fight COVID-19 [13]. Compared to the traditional imaging workflow that heavily relies on human labor, AI enables more safe, accurate, and efficient imaging solutions. Recent AI-empowered applications in COVID-19 mainly include the dedicated imaging platform, the lung and infection region segmentation, the clinical assessment and diagnosis, as well as the pioneering basic and clinical research. Moreover, many commercial products have been developed, which successfully integrate AI to combat COVID-19 and clearly demonstrate the capability of the technology. The Medical Imaging Computing Seminar (MICS) 1, a China’s leading alliance of medical imaging scholars and start-up companies, organized this first online seminar on COVID-19 on February 18, 2020, which attracted more than ten thousands of visits. All the above examples show the tremendous enthusiasm cast by the public for AI-empowered progress in the medical imaging field, especially during the ongoing pandemic.
Due to the importance of AI in all the spectrum of the imaging-based analysis of COVID-19, this review aims to extensively discuss the role of medical imaging, especially empowered by AI, in fighting the COVID19, which will inspire future practical applications and methodological research. In the following, we first introduce intelligent imaging platforms for COVID-19 and then summarize popular machine learning methods in the imaging workflow, including segmentation, diagnosis, and prognosis. Several publicly available datasets are also introduced. Finally, we discuss several open problems and challenges. We expect to provide guidance for researchers and radiologists through this review. Note that we review the most related medical-imaging-based COVID-19 studies up to March 31, 2020.
- AI-EMPOWERED CONTACTLESS IMAGING WORKFLOWS
Healthcare practitioners are particularly vulnerable concerning the high risk of occupational viral exposure. Imaging specialists and technicians are of high priority, such that any potential contact with the virus could be under control. In addition to the personal protective equipment (PPE), one may consider dedicated imaging facilities and workflows, which are significantly important to reduce the risks and save lives.
- Conventional Imaging Workflow
Chest X-ray and CT are widely used in the screening and diagnosis of COVID-19 [7-12]. It is important to employ a contactless and automated image acquisition workflow to avoid the severe risks of infection during the COVID-19 pandemic. However, the conventional imaging workflow includes inevitable contact between technicians and patients. Especially, inpatient positioning, technicians first assist in posing the patient according to a given protocol, such as head-first versus feet-first, and supine versus prone in CT, followed by visually identifying the target body part location on the patient and manually adjusting the relative position and pose between the patient and the X-ray tube. This process puts the technicians in close contact with the patients, which leads to high risks of viral exposure. Thus, a contactless and automated imaging workflow is needed to minimize the contact.
- AI-Empowered Imaging Workflow
Many modern X-ray and CT systems are equipped with cameras for patient monitoring purposes [14-17]. During the outbreak of COVID-19, those devices facilitate the implementation of a contactless scanning workflow. Technicians can monitor the patient from the control room via a live video stream from the camera. However, from only the overhead view of the camera, it is still challenging for the technician to determine the scanning parameters such as the scan range. In this case, AI is able to automate the process [18-26] by identifying the pose and shape of the patient from the data acquired with visual sensors such as RGB, Time-of-Flight (TOF) pressure imaging [27] or thermal (FIR) cameras. Thus, the optimal scanning parameters can be determined.
One typical scanning parameter that can be estimated with AI-empowered visual sensors is the scan range that defines the starting and ending positions of the CT scan. Scan range can be identified by detecting anatomical joints of the subject from the images. Much recent work [28-30] has focused on estimating the 2D [31-36] or 3D keypoint locations [29, 37-40] on the patient body. These keypoint locations usually include major joints such as the neck, shoulders, elbows, ankles, wrists, and knees. Wang et al. [41] have shown that such an automated workflow can significantly improve scanning efficiency and reduce unnecessary radiation exposure. However, such key points usually represent only a very sparse sampling of the full 3D mesh [42] in the 3D space (that defines the digital human body).
Other important scanning parameters can be inferred by AI, including ISO-centering. ISO-centering refers to aligning the target body region of the subject so that the center of the target body region overlaps with the scanner ISO center and thus the overall imaging quality is optimal. Studies have shown that, with better ISO-centering, radiation dosage can be reduced while maintaining similar imaging quality [43]. In order to align the target body region to the ISO center, and given that anatomical keypoints usually represent only a very sparse sampling of the full 3D mesh in the 3D space (defining the digital human body), Georgakis et al. [44] propose to recover human mesh from a single monocular RGB image using a parametric human model SMPL [45]. Unlike other related studies [46], they employ hierarchical kinematic reasoning for each kinematic chain of the patient to iteratively refine the estimation of each anatomical key point to improve the system’s robustness to clutters and partial occlusions around the joints of the patient. Singh et al. [19] present a technique, using depth sensor data, to retrieve a full 3D patient mesh by fitting the depth data to a parametric human mesh model based on anatomical landmarks detected from RGB image. One recent solution proposed by Ren et al. [42] learns a model that can be trained just once and have the capability to be applied across multiple such applications based on dynamic multi-model inference.
With this framework in application with an RGB-depth input sensor, even if one of the sensor modalities fails, the model above can still perform 3D patient body inference with the remaining data.
- Applications in COVID-19
During the outbreak of COVID-19, several essential contactless imaging workflows were established[18, 41, 42], from the utilization of monitoring cameras in the scan room [14-16, 28], or on the device [47], to mobile CT platforms [18, 47-50] with better access to patients and flexible installation.
A notable example is an automated scanning workflow based on a mobile CT platform empowered by visual AI technologies [18]. The mobile platform is fully self-contained with an AI-based pre-scan and diagnosis system [47]. It was redesigned into a fully isolated scan room and control room. Each room has its own entrance to avoid any unnecessary interaction between technicians and patients.
After entering the scan room, the patient is instructed, by visual and audio prompts, to pose on the patient bed. Technicians can observe through the window and also the live video transmitted from the ceiling-mounted AI camera in the scan room, and correct the pose of the patient if necessary. Once the patient is deemed ready, either by the technician or the motion analysis algorithm, the patient positioning algorithm will automatically recover the 3D pose and fully-reconstructed the mesh of the patient from the images captured with the camera [42]. Based on the 3D mesh, both the scan range and the 3D centerline of the target body part of the patient are estimated and converted into control signals and optimized scanning parameters for the technician to verify. If necessary, the technician can make adjustments. Once verified, the patient bed will be automatically aligned to ISO center and moved into CT gantry for scanning. After CT images are acquired, they will be processed and analyzed for screening and diagnosis purposes… .
Conclusions
COVID-19 is a disease that has spread all over the world. Intelligent medical imaging has played an important role in fighting against COVID-19. This paper discusses how AI provides safe, accurate, and efficient imaging solutions in COVID-19 applications. The intelligent imaging platforms, clinical diagnosis, and pioneering research are reviewed in detail, which covers the entire pipeline of AI-empowered imaging applications in COVID-19. Two imaging modalities, i.e., X-ray and CT, are used to demonstrates the effectiveness of AI-empowered medical imaging for COVID-19.
It is worth noting that imaging only provides partial information about patients with COVID-19. Thus, it is important to combine imaging data with both clinical manifestations and laboratory examination results to help better screening, detection, and diagnosis of COVID-19. In this case, we believe AI will demonstrate its natural capability infusing information from these multi-source data, for performing accurate and efficient diagnosis, analysis, and follow-up.
About KSRA
The Kavian Scientific Research Association (KSRA) is a non-profit research organization to provide research / educational services in December 2013. The members of the community had formed a virtual group on the Viber social network. The core of the Kavian Scientific Association was formed with these members as founders. These individuals, led by Professor Siavosh Kaviani, decided to launch a scientific / research association with an emphasis on education.
KSRA research association, as a non-profit research firm, is committed to providing research services in the field of knowledge. The main beneficiaries of this association are public or private knowledge-based companies, students, researchers, researchers, professors, universities, and industrial and semi-industrial centers around the world.
Our main services Based on Education for all Spectrum people in the world. We want to make an integration between researches and educations. We believe education is the main right of Human beings. So our services should be concentrated on inclusive education.
The KSRA team partners with local under-served communities around the world to improve the access to and quality of knowledge based on education, amplify and augment learning programs where they exist, and create new opportunities for e-learning where traditional education systems are lacking or non-existent.
FULL Paper PDF file:
Review of Artificial Intelligence Techniques in Imaging Data Acquisition, Segmentation and Diagnosis for COVID-19Bibliography
Authors
F. Shi, J. Wang, J. Shi, Z. Wu, Q. Wang, Z. Tang, K. He, Y. Shi, D. Shen
Year
2020
Title
Review of Artificial Intelligence Techniques in Imaging Data Acquisition, Segmentation, and Diagnosis for COVID-19
Journal
IEEE Reviews in Biomedical Engineering
PDF reference and original file: Click here
Somayeh Nosrati was born in 1982 in Tehran. She holds a Master's degree in artificial intelligence from Khatam University of Tehran.
