Introduction
The successes in the 2000s and 2010s in developing reactive machine AI and limited-memory AI methods to detect and assess the present and severity of x-ray chest CT imaging findings associated with COVID-19 pneumonia, COPD, ILD, and lung cancer have spurred the development of multiple quantitative CT (QCT) lung AI companies, such as VIDA, that offer point-of-care lung CT AI products to assess lung diseases. VIDA’s specialized FDA-approved lung CT AI program, VIDA Insights v3.0, can be run independently or inside a larger medical imaging AI ecosystem. It is now possible for every chest CT scan to be analyzed in near real time for QCT AI metrics of COVID-19 pneumonia, COPD, ILD, and lung nodules. The lung CT AI information is automatically generated and inserted in imaging physicians’ reports and have an immediate impact on the detection and assessment of the severity of diffuse lung disease and lung cancer. It has been a long journey from the mid-1970s to the early 2020 s, but QCT AI of lung disease for the clinical care and treatment of COPD and ILD is now coming of age.
Healthcare Imaging IT
Imaging technology in modern healthcare systems relies on an ecosystem of AI agents to deliver high-quality medical care to patients. The AI agents covered in this chapter include the electronic medical record, radiology information system, picture archiving and communication, voice recognition and reporting, and disease-specific quantitative lung CT AI agents.
Electronic Medical Record (EMR)
The EMR is an AI program that stores, transmits, and displays critical medical information for a patient seen within a healthcare organization (e.g., clinic, hospital, web). The EMR is used by medical personnel to care for patients. The EMR is interfaced with other AI programs that perform more specialized AI functions, such as medical imaging. The medical imaging AI programs include picture archiving and communication (PACS) software programs, radiology information system (RIS) software programs, and voice recognition and reporting (VR) software programs. The PACS, RIS, and VR programs all work together to enable imaging physicians to create imaging reports on the medical imaging studies contained in the PACS and then have these same imaging reports sent to the EMR where they can be viewed by the ordering physician and the other healthcare workers and the patient.
Picture Archiving and Communication System (PACS)
The PACS is the core technology that is responsible for storing, transmitting, and displaying medical images within healthcare systems. The major PACS components include hardware, software, and local area networks ( Fig. 9.1 ). The hardware consists of computer servers to store and transmit medical imaging studies, computer workstations where the medical images are interpreted by imaging physicians, and local area networks that connect the medical imaging devices (e.g., x-ray CT scanners) to the computer servers that store the medical imaging data. The PACS software components include the software that controls the storage and transmission of medical images, software to display and interact with the medical images, and software that exchanges information with the RIS and EMR.
The data format for medical images has been standardized using what is called the Digital Imaging and Communications in Medicine (DICOM) standard. DICOM provides standard protocols for exchanging and storing medical image data including both image data and text associated with the image data. Manufacturers of PACS software have adopted the DICOM standard. The Medical and Imaging Technical Alliance (MITA) division of the National Electrical Manufacturers Association (NEMA) manages a structured document describing the DICOM standard. The DICOM Standard’s document currently includes 20 Parts, Parts 1–8, Parts 10–12, and Parts 14–22. Each of the DICOM Standard Parts address specific subject areas, for example Part 10 addresses “Media Storage and File Format for Media Interchange”. The International Organization for Standardization has recognized DICOM as the ISO 10252 standard.
The development of web-based PACS software has driven the need to develop an application programming interface (API) for handling DICOM-compliant medical imaging data on the worldwide web. DICOMweb has been developed to provide an API to support DICOM standards for web applications similar to what DICOM standards have done for PACS systems. The DICOMweb API provides software programming standards for web-based medical imaging applications for sending, retrieving, and querying images and image-related information.
Current PACS systems provide for “hanging protocols” to be assigned to each type of imaging study. Hanging protocols are implemented within the DICOM standard. The term “hanging” comes from the historic method of hanging hardcopy film on a viewer for the imaging physician to visually interpret and report the findings of a given imaging study. The PACS hanging protocols provide powerful tools to organize the display of the different imaging studies, such as chest CT scans. The hanging protocols are not specific to a patient but are specific to the type of imaging study performed. The chest CT hanging protocol can include multiple parameters including anatomic laterality (right versus left), the anatomic plane that is displayed (e.g., axial, coronal, sagittal), reconstruction method (FBP versus IR), reconstruction kernel, slice thickness, slice interval, window width (WW), window level (WL), location on the displays of the current chest CT images, and historic chest CT’s if they exist. A lung CT AI program that produces additional DICOM outputs needs to be able to integrate into the hanging protocol of the different PACS vendor systems. This is usually done by generating a new CT series that can be handled by the PACS like any other CT imaging series.
Radiology Information Software (RIS)
The RIS is a medical information AI agent that communicates through the medical imaging LAN network with the EHR to send and receive patient-specific medical imaging study information. The RIS has recently been incorporated into the EMR rather than continue to be a stand-alone software program. The RIS typically has imaging modality and subspecialty-specific worklists that inform the imaging physician and the imaging technologist as to which patients need a specific imaging study (e.g., chest CT) and why the specific imaging study was ordered. The RIS has access to patient-specific information stored in the EHR to inform the imaging physician as to why the medical imaging examination was ordered (e.g., patient chief complaint and symptoms), and to receive the structured reporting documents generated by the imaging physician, usually using VR reporting software. The RIS software sends the report on the medical imaging study back to the EMR to update the patient’s electronic medical record.
The communication standards between the EMR, RIS, and PACS are implemented using the HL7 standard. Health Level Seven International (HL7) was founded in 1987. HL7 is a non-profit organization that develops ANSI-accredited framework and standards for the exchange, integration, and sharing of electronic health information. The HL7 standard provides a mechanism for the electronic exchange of alphanumeric medical information (e.g., text and numbers) between the different software programs that are running on a medical informatics local area network (LAN), including the EMR, RIS, PACS, and other medical informatics programs in the healthcare IT environment, such as the Laboratory Information System (LIS).
Medical Imaging Reporting and Voice Recognition Software (VR)
Every medical imaging study needs to be visually assessed and an accurate and concise report generated for the imaging study. Voice recognition (VR) software has been used in reporting medical imaging studies over the past couple of decades. The currently available versions of the FDA-approved VR reporting software are quite capable and in use by most imaging physicians. The most popular VR reporting software for imaging studies including CT is Nuance’s AI-driven PowerScribe 360 and PowerScribe One.
Clinical Lung CT AI Software
The development of clinical lung CT AI software to detect and assess lung disease must be able to fit into the larger medical imaging IT environment previously described. The goal of clinical lung CT AI software should be to accurately assess chest CT studies for quantitative metrics of lung disease while integrating smoothly into the clinical imaging workflow. This means that the lung CT AI software needs to access chest CT imaging data either directly from the x-ray CT scanner or the PACS. The lung CT AI software needs to run quickly in the background so that the results are available when the chest CT study is interpreted by the imaging physician. The outputs from the lung CT AI software need to be DICOM compliant for results that are stored in the PACS with the chest CT study that has been analyzed. The HL7-compliant alpha-numeric outputs from the lung CT AI program, such as LAA −950 for the assessment of emphysema, should be able to flow into the VR reports along with other structured text reporting fields.
VIDA Insights–Clinical Lung CT AI Software
As of December 2021, VIDA Insights has two modules: Density/tMPR and Texture/Subpleural View. VIDA lung CT AI software is rapidly evolving, and the latest information on VIDA lung CT AI can be accessed on the web. VIDA Insights can be integrated with the CT scanner PACS ( Fig. 9.2 ).
VIDA Insights assesses each of the imaging series obtained for the chest CT study and finds the study with the best technique for assessing the lungs. This includes looking for a contiguous data acquisition with a slice thickness of 1.5 mm or less and FBP reconstruction with a neutral kernel (see Chapter 3 ). Then, the selected series is automatically analyzed by a deep learning AI software program to see if there are motion artifacts within the lungs. VIDA Insights will use the results of these quality control steps to inform the imaging physician as to the acceptability of the chest CT in question. If there are severe quality issues, then the chest CT will not be analyzed.
VIDA Insights automatically segments the lungs and airways from the rest of the thoracic anatomy using a deep learning AI software program. The lung segmentation is robust and includes segmenting the individual lungs and lobes. The challenge of separating the lungs from the mediastinal and chest wall structures automatically and accurately is substantial, since a peripheral area of consolidation from COVID-19 pneumonia that abuts the pleura or diaphragm has similar tissue densities. Robust lung CT image segmentation is an ongoing research-and-development process pursued through public and industry-funded research.
The design of VIDA Insights was modeled after the method a thoracic radiologist applies to the visual interpretation of the lung findings on a chest CT study. This includes finding the best CT image series to assess the lungs and checking that the chest CT scan was done at the correct lung volume, with the correct chest CT scan protocol, and ensuring there are no motion artifacts that will adversely affect the visual interpretation of the lung CT images. Then, the lungs are assessed for evidence of increased, normal, or decreased lung volumes. Increased lung volumes are associated with obstructive airway diseases, such as COPD and asthma. Decreased lung volumes are associated with restrictive lung diseases, such as ILD. Normal or near-normal lung volumes are associated with acute lung disease, such as COVID-19 pneumonia. Next, the lung CT images are assessed for areas of decreased density due to emphysema, or increased density due to pneumonia or pulmonary fibrosis. Then, the lung CT images are assessed for characteristic texture patterns of diseased lung tissue: ground-glass/reticular opacities, consolidation, and honeycombing.
VIDA Insights Density/tMPR Reactive Machine AI Tool for Assessing Volumes, LAA, and HAA
Chapter 5 discussed the COPDGene study use of four novel criteria to diagnose COPD and to assess the impact of these criteria on COPD progression and mortality: “Exposure, Symptoms, CT Structural Abnormality, Spirometry”. Exposure was defined as having a 10-pack-year or greater smoking history Symptoms were defined as self-reporting a modified Medical Council (mMRC) dyspnea score of 2 or greater and/or chronic bronchitis (self-reported chronic cough and phlegm). CT structural abnormality was defined as having one or more of the following: LAA −950 equal to 5% or greater on a TLC chest CT scan (QCT structural measure of emphysema), LAA −856 ≥15% on expiratory CT scan (QCT functional measure of air trapping), Pi10 equal to 2.5 mm or greater (QCT structural measure of airway wall thickening). The results of this study provide guidance and value to assessing the LAA −950 to detect and assess emphysema in a clinical lung CT AI program.
VIDA Insights Density/tMPR assessment uses a reactive machine AI approach to analyze the lung voxel histogram curve obtained from each lung and lobe from the segmented lung CT images. The volume of each lung and lobe can be calculated by knowing the volume of each lung CT voxel and the number of lung CT voxels in each lung and lobe. The lung and lobe volumes are calculated by multiplying the volume of each voxel by the number of voxels in each lobe and lung, and the volume is expressed in liters for each lung and lobe.
VIDA Insights Density/tMPR then analyzes the lung CT voxel histogram curve to calculate the number of voxels <−950 HU or the LAA −950 discussed in Chapter 5 . The LAA −950 is a QCT metric of emphysema and when the LAA −950 is 5% or greater, the affected lung or lobe is highlighted in yellow ( Figs. 9.3 and 9.4 ). As the percentage of LAA −950 increases above 5%, the color deepens in the affected lung and lobe.
