Advice
The technology
The technology
The chest imaging artificial intelligence (AI) technologies in this briefing are standalone software platforms that use machine or deep learning algorithms to analyse or interpret radiology images. Some technologies allow images to be transferred from the hospital to the software platform, which is hosted in an NHS accredited secure data centre. The software analyses the chest DICOM (digital imaging and communications in medicine) image using proprietary algorithms. The image analysis may be sent directly back to the hospital to be viewed with hospital systems such as picture archiving and communication system (PACS) and some radiology information systems using protocols such as DICOM and HL7. Some technologies may also allow uploading and viewing of images and analysis using a web interface.
Version updates and periodic maintenance activities are needed for these technologies. This can be done remotely.
The technology may help identify images as normal or abnormal, highlight suspected abnormalities and provide results as heat maps or clinically relevant labels. It may also provide support for prioritising CTs for specialist review. The AI analyses are intended to be used with radiology images to support radiologist review and decision to improve diagnostic accuracy. Turnaround time may be decreased for time sensitive conditions such as pneumothorax or catheter malposition. They are not intended to be used as medical advice.
The following technologies are post-processing image analysis software for chest CT. Other, similar technologies may be available but are not included in this briefing (for example if they were not identified, or the company chose not to participate).
-
Veye Chest (Aidence) – used for automatic detection, classification, measurement and growth assessment of solid and sub-solid pulmonary nodules. Can be used on low-dose or standard-dose, and non-contrast or post-contrast scans with a maximum axial slice thickness of 3 mm.
-
icolung (icometrix) – used for automatic detection, segmentation and measurement of lung abnormalities in 5 lung lobes in non-contrast scans. Has a maximum axial slice thickness of 5 mm (about 1 mm thickness is recommended). The company states that version 0.6.0 onwards is expected to provide support for COVID-19 diagnosis by including the probability of an image being from a person with COVID-19.
-
Veolity (MeVis) – used for reading chest CTs including automatic detection, segmentation and measurement of pulmonary nodules. Can be used on low-dose or standard-dose, and non-contrast or post-contrast scans. The algorithm automatically compares the current scan with previous scans and assesses change in nodule size.
Innovations
AI for analysing chest CT may help increase diagnostic accuracy and reduce time to diagnosis by providing additional information for radiologists. The technology automatically reads medical images and identifies abnormalities.
Current care pathway
Depending on the intended population, the technology potentially applies to a range of NICE guidance including:
-
The section on diagnosis and staging in NICE's guideline on lung cancer advises to offer people with known or suspected lung cancer a contrast-enhanced chest CT scan to further the diagnosis and stage the disease. It states to include the liver, adrenals and lower neck in the scan.
-
The section on diagnosis in NICE's guideline on metastatic malignant disease of unknown primary origin in adults states to offer a CT scan of the chest, abdomen and pelvis.
-
NICE's guideline on venous thromboembolic diseases advises to offer patients an immediate CT pulmonary angiogram if pulmonary embolism is suspected along with a likely 2-level Wells score.
-
NICE's guideline on diagnosing chronic obstructive pulmonary disease (X-ray images or CT scans).
-
NICE's guideline on idiopathic pulmonary fibrosis in adults (CT and X-ray).
-
NICE's guideline on tuberculosis recommends that local hospitals, Clinical Commissioning Groups and local multidisciplinary teams should consider developing a pathway for people with imaging highly suspected of tuberculosis. X-ray imaging or CT may be used in children and young people with suspected pulmonary tuberculosis, or in a number of groups with suspected extrapulmonary tuberculosis.
-
Also see the NICE topic page on respiratory conditions.
Population, setting and intended user
Radiologists (and also radiology specialist registrars and reporting radiographers) in secondary care review and interpret images from people referred for radiological imaging because of suspected abnormalities in the chest. This allows them to make diagnoses and inform planning of patient management. AI for chest CT is intended to support this process by providing an additional source of automatic analysis.
Training may be needed for radiologists to learn how to use the software and the reports it produces.
Costs
Technology costs
-
Veye Chest (Aidence), price per output ranges from £5 to £7.50, depending on features selected and volumes of scans. A yearly fee of between £4,000 and £9,000 for cloud server hosting, monitoring and support. IT integration, training and deployment is a one-off cost of £8,500.
-
icolung (icometrix) is offered pro bono as part of the icovid.ai initiative. The costs associated with consumables, maintenance or training are included free of charge during the endemic phase of the COVID pandemic, and at least for 6 months. Costs may apply after this phase.
-
Veolity (MeVis) offers a one-off perpetual software licence for £44,000 with no per scan costs. This includes first year maintenance (yearly maintenance cost is then £8,800). Initial installation, testing and training costs are £9,000. Yearly support costs are £6,000.
Resource consequences
No published evidence on the resource consequences of AI for chest CT was found.
The costs of adopting chest CT reporting services vary among the technologies included in this review. The cost associated with the technologies generally consists of integration costs, fixed cost per scan processed and yearly maintenance costs. Chest CT is used in the NHS to help diagnosis in a number of clinical pathways. The unit cost of the technologies depends on the total CT throughput.
Minimal changes are needed in facilities or infrastructure as long as software companies comply with NHS communication standards and there is a robust IT infrastructure in the implementing organisation. Existing IT infrastructure and software may vary across NHS organisations, and unforeseen issues may arise because of implementing novel software.