The technology

Diagnostic artificial intelligence (AI) software has been developed to review and report abnormalities in CT brain scans. These AI packages have automated analysis of CT brain scans, including non-contrast CT (NCCT), CT angiography (CTA) and CT perfusion (CTP) imaging. In most cases, the software aids detection and prioritisation of critical cases, such as intracranial haemorrhage and large vessel occlusion in stroke.

The AI software packages in this briefing are designed to automatically detect and notify healthcare professionals of abnormalities after analysis of brain CT scans. The software is designed to integrate with all standard imaging systems. They have automated patient prioritisation and alert systems for critical cases. Some can populate radiology reports with preliminary findings. The technologies are for use in addition to standard care. Two packages assess CT images for abnormalities because of trauma, dementia and stroke. The remaining 5 identify large vessel occlusion or intracranial haemorrhage in stroke. The packages are:

  • Aidoc: head (Aidoc). Programmes for detecting intracranial haemorrhage and hyperdensities after NCCT imaging and for large vessel occlusion after CTA imaging. The software also has automated patient prioritisation and a real-time alert system. It integrates into current imaging systems and results can be viewed as digital imaging and communication in medicine (DICOM) output images.

  • e‑CTA and e‑ASPECTS (Brainomix). e‑ASPECTS analyses NCCT scans for hypodensity and generates a probability map of regional ischaemic change, the volume of this change, and an automated ASPECT score. e‑CTA is designed to detect large vessel occlusion location and standardised assessments of collateral scores after CTA imaging. Both applications help identify people eligible for thrombectomy or thrombolysis. The software integrates with current imaging systems and results can be viewed as visual reports through DICOM output images, email notifications and a web browser.

  • Icobrain (Icometrix). This quantifies and reports the volume of relevant brain structures related to dementia, stroke and traumatic brain injury. In traumatic brain injury the software quantifies: epidural, subdural and intraparenchymal lesions; midline shift; left, right lateral ventricles and fourth ventricle. In stroke, core and penumbra sections of the brain are assessed, and in dementia the whole brain volume and lateral ventricles. The software integrates with current imaging systems and results can be viewed as visual reports through DICOM output images, email notifications and a web browser.

  • qER (Qure). qER detects and quantifies a range of brain abnormalities after NCCT imaging, and populates a radiology reporting template with preliminary findings, patient prioritisation and alert systems including mobile notifications. Brain pathologies identified by qER include intracerebral bleeds and their subtypes, infarcts, mass effect, midline shift and cranial fractures. It integrates with current imaging systems.

  • Zebra triage (Zebra Medical Vision). This detects and annotates intracranial haemorrhage after NCCT imaging and automates patient prioritisation and a real-time alert system. It integrates with the current imaging worklist and viewer with an accompanying alert widget.

  • DLCExpert (Mirada Medical) has not been included in this briefing because of differences in indication and patient population. This technology uses AI to help treatment planning for radiation therapy (including head and neck cancers). It uses deep learning algorithms for automated contouring of organs at risk and anatomical structures. The company claims the algorithms have been trained on clinical examples and validated against consensus guidelines, such as the European Society for Radiotherapy and Oncology head and neck guidelines (ESTRO) for delineation of organs at risk.

Innovations

The software packages use AI to automatically analyse CT brain scans for abnormalities, alert radiologists to critical cases, and prioritise cases. Some software systems also report preliminary findings, and some assess brain structures or abnormalities. Companies claim these advances in technology will result in time-sensitive cases being reviewed more quickly, meaning faster treatment and improved patient outcomes. Companies also claim reporting preliminary findings reduces reading and dictation time for clinicians and prevents subtle abnormalities being missed.

Current care pathway

The diagnosis of a suspected brain abnormality is usually confirmed by a radiologist after a review of brain imaging techniques, usually CT scans. Results from CT brain scans are typically available in a few days to a week, depending on the urgency of the case. However, in emergency situations reports are usually available within 24 hours, and urgent cases are often reviewed within a couple of hours.

For people admitted with suspected stroke, NICE's guideline on stroke and transient ischaemic attack in over 16s: diagnosis and initial management recommends the prompt use of a validated screening tool, FAST, to assess people with sudden onset neurological symptoms. On admission the validated tool, ROSIER, is used to diagnose stroke or transient ischaemic attack. When transient ischaemic attack is suspected, people are given 300 mg aspirin and referred for specialist assessment. Brain imaging is not recommended. People admitted with suspected acute stroke are referred to a specialist stroke unit. An NCCT is recommended as soon as possible and within 24 hours. People with an increased risk should be scanned immediately. When acute ischaemic stroke is suspected and symptom onset is more than 6 hours before, CTA or CTP is done. Stroke cases are time-sensitive and treatment is needed as soon as possible. The treatment decision depends on the outcome of the imaging and the time since symptom onset. Ischaemic causes of stroke need immediate treatment with aspirin or anticoagulants; haemorrhagic causes of stroke need anticoagulation reversal. NICE's interventional procedures guidance for mechanical clot retrieval for treating acute ischaemic stroke recommends thrombectomy to treat ischaemic stroke for people who are eligible after CTA or CTP imaging. For people who need a thrombectomy, the procedure should be done before admission to a specialist stroke unit.

For the diagnosis of dementia, the NICE guideline on dementia: assessment, management and support for people living with dementia and their carers recommends initial cognitive and physical assessments, and taking blood and urine samples. A person is referred to the specialist dementia diagnostic service where validated criteria guide diagnosis. If Alzheimer's disease is suspected but not confirmed, fluorodeoxyglucose positron emission tomography (FDG-PET) or cerebrospinal fluid testing is recommended to inform the diagnosis. People diagnosed are offered interventions to promote cognitive function.

Population, setting and intended user

This technology is for people with suspected brain abnormalities. Most of the technologies described in this briefing are for people with a suspected intracranial haemorrhage or acute ischaemic stroke. Some are also designed to detect abnormalities related to traumatic brain injury and dementia.

The technology is used by radiologists and neuroradiologists in imaging facilities in secondary and tertiary care settings as a decision support tool.

Costs

Technology costs

  • Aidoc: head. The licence cost of the technology ranges from £25,000 to £60,000 per year depending on the volume of exams and different workflow requirements. The cost covers the software for intracranial hyperdensities and large vessel occlusion as well as all associated costs, for example, training and maintenance.

  • e‑ASPECTS and e‑CTA. The licence cost of the technology is £30,000 per year excluding VAT for unlimited stroke patients admitted to comprehensive stroke centres. The cost is reduced to £10,000 per year for primary strokes centres doing thrombolysis only if purchased as part of a full stroke network licence.

  • Icobrain Ix. Pay-per-use models start at £45, with reductions for increased volumes. Subscription models range from £8,250 for small hospitals (including 300 analyses) to £66,000 for large hospitals (including 2,400 analyses and services).

  • qER. The cost for the technology as it has been described is £25,000.

  • Zebra triage. The technology licence costs between £40,000 and £80,000 depending on the size of the NHS trust. These costs cover an 'all-in-one' bundle, including 5 AI algorithms for detecting intracranial haemorrhage, pneumothorax, pleural effusion, vertebral fracture and screening for breast cancer.

Costs of standard care

According to the national tariff payment system 2019/2020, a routine CT scan for 1 area costs £69 without contrast. A CT scan of 1 area with imaging before and after contrast costs £90. This includes the cost of reporting. When another healthcare trust is needed to review the CT scan there is an additional cost of £20.

Resource consequences

These technologies would typically cost more than standard care but may result in cost savings related to reduced radiologist's time in reviewing and reporting CT brain scans. Prioritisation of critical cases may also reduce complications related to delayed treatment. Radiologists would need training to ensure appropriate use of the technology.