GLOSSARY

Computer-Aided Diagnosis Software

Computer-aided diagnosis (CAD) software is a system — often AI-powered in modern implementations — that analyses medical images and produces objective, quantitative outputs — measurements, classifications, or probability scores — to support clinical decision-making or serve as analytically validated endpoints in clinical trials. It does not replace the clinician; it makes assessment faster, more consistent, and less dependent on individual reader variability.

 

What is computer-aided diagnosis software?

Computer-aided diagnosis software — also written as computer-aided diagnosis or referred to as CAD software, CADx, or AI-assisted diagnosis — is a system that analyses medical images and provides quantitative measurements, classifications, or probability scores to support clinical decision-making. Modern CAD systems are often AI-powered, particularly those using deep learning for image analysis, though the term historically also encompassed rule-based and classical statistical approaches.

CAD software trained on large, labelled imaging datasets learns to detect and characterise features that are difficult to quantify reliably by visual inspection alone. The output is presented to the reviewing clinician as supporting data alongside the image itself.

In clinical care, CAD is one input among many informing a decision about an individual patient. In clinical trials, CAD outputs serve a more specific function: they produce objective, reproducible endpoint measurements that can be compared across patients, sites, and time points without the inter-reader variability that affects manual assessment.

CADe, CADx, and CADt — the three types

Not all CAD software does the same thing. The distinction matters for clinical procurement, trial design, and regulatory classification.

CADe — computer-aided detection

Identifies and flags potential abnormalities for the reviewing clinician. Answers the question: is there something here that warrants attention?

CADe highlights regions of interest — a possible lesion, a suspicious nodule — but does not characterise, measure, or classify what it has found. This is the most common type of CAD software in radiology screening workflows, where the goal is to reduce miss rates on high-volume reads.

CADx — computer-aided diagnosis

Goes further. Characterises or classifies the abnormality, providing quantitative measurements, probability scores, or severity ratings. Answers the question: what is this finding, and how significant is it?

Lesion volume measurement in multiple sclerosis, brain atrophy quantification in Alzheimer's disease, and amyloid burden scoring in PET imaging are all CADx applications. CADx is the category most directly relevant to clinical trials, where quantitative characterisation — not merely detection — is required for endpoint assessment.

Most of QMENTA's imaging algorithm catalog provides CADx outputs: volumetric brain measurements, lesion characterisation, and quantitative biomarker values. CADx systems often face more stringent validation requirements due to their role in quantitative assessment and decision support — though FDA classification ultimately depends on intended use and risk classification rather than the CADe/CADx label alone.

CADt — computer-aided triage

Prioritises imaging worklists based on urgency, routing the most time-critical cases to the top of the radiologist's queue. Primarily a radiology workflow efficiency tool with limited direct relevance to clinical trial endpoint assessment.

Computer-aided diagnosis software vs. a radiologist

Under FDA and EMA regulatory frameworks, most current CAD software does not replace the radiologist — it functions as a decision support tool.

FDA-cleared CAD software is presented to the reviewing clinician as supporting data. The clinician remains fully responsible for the final diagnosis and any treatment decisions. This is not a legal technicality — it reflects how CAD software actually performs. Current AI systems excel at specific, well-defined measurement tasks: counting lesions, measuring volumes, flagging signal anomalies. They do not replicate the broader clinical reasoning a radiologist applies when integrating imaging findings with a patient's history, symptoms, and differential diagnosis.

In neuroimaging clinical trials, CAD software is most valuable not as a replacement for expert reading but as a consistency layer — ensuring that every scan across every site in a multi-year trial is processed with the same algorithm, the same parameters, and the same version, regardless of which radiologist reviews the output.

FDA clearance and regulatory classification

Computer-aided diagnosis software intended for clinical use is classified as a Software as a Medical Device (SaMD) under FDA and CE Mark frameworks.¹

Many CAD systems for radiology are cleared via FDA 510(k) — a premarket notification demonstrating substantial equivalence to a legally marketed predicate device — before clinical use in the United States. Some tools use the De Novo pathway or, for higher-risk devices, the PMA pathway, depending on intended use and risk classification.

The FDA's Digital Health Center of Excellence provides regulatory guidance specific to AI and machine learning-based SaMD, including requirements for analytical validation, algorithm transparency, and post-market performance monitoring.²

QMENTA's Care Platform received 510(k) clearance from the FDA in November 2021, clearance number K202718, authorising use by qualified healthcare providers to manage imaging data and apply AI-powered analysis tools for neurological evaluation.³ This clearance covers the platform's intended use — including data management and the application of certain analysis tools — rather than individual clearance of every algorithm in the catalog. Sponsors should confirm the regulatory status of specific algorithms for their intended clinical or trial use.

For clinical trials, FDA clearance is not required to use CAD software. The relevant standard is analytical validation, not clearance status. CAD-derived measurements can be used as primary or secondary efficacy endpoints in a regulatory submission without the software holding a 510(k) clearance — provided it is analytically validated for the specific intended use.

In practice, analytical validation means demonstrating reproducibility and accuracy of the software's outputs across the scanner types, imaging protocols, sites, and patient populations represented in the trial. Clearance is useful supporting evidence — it signals the software has met baseline regulatory standards — but it does not substitute for endpoint-specific validation and does not automatically qualify a tool for trial endpoint use.

Sponsors should define the validation strategy for any CAD-derived endpoints during protocol development and align that strategy with the relevant regulatory agency before trial initiation.

 

Using CAD-derived measurements as trial endpoints?
QMENTA's imaging scientists can advise on analytical validation requirements, algorithm selection, and how to document CAD endpoint methodology for regulatory submissions.
Talk to an expert →

 

Computer-aided diagnosis in neuroimaging

Neuroimaging is one of the most mature and rapidly evolving areas for CAD software, for three reasons: the brain's anatomy is complex and highly variable across individuals; neurological diseases typically progress slowly over years, requiring sensitive longitudinal measurement; and unlike most other organs, the brain cannot be repeatedly biopsied, making imaging the primary window onto disease progression.

Key CADx applications in current neurological clinical trials:

White matter lesion detection and volume measurement
Used in multiple sclerosis trials to quantify disease activity and measure treatment response to disease-modifying therapies. Inter-reader variability in manual MS lesion counting has been reported in some studies at approximately 20–30% between experienced neuroradiologists reading the same scan. AI-assisted measurement reduces this variability at the measurement level, improving consistency across readers and time points.⁴

White matter microstructure assessment — diffusion tensor imaging (DTI)
Quantifies axonal and myelin integrity through fractional anisotropy (FA) and mean diffusivity (MD), detecting microstructural damage in tissue that appears normal on conventional MRI. Used in TBI trials as a prognostic endpoint — the CENTER-TBI study demonstrated DTI metrics significantly improved outcome prediction in mild TBI patients with normal CT scans, and highlighted its value for enriching trial populations.5 In MS, DTI captures white matter tract degeneration in patients with progression independent of relapse activity (PIRA), beyond what lesion volume measures alone detect.6 DTI is also increasingly used in trials for rare white matter disorders — including leukodystrophies and adrenoleukodystrophy — where tract-level microstructural change is often the most sensitive available measure of disease progression.

Brain atrophy quantification
Measured using tools including SIENAX and FreeSurfer, used across Alzheimer's disease, Parkinson's disease, and MS trials as a sensitive marker of neurodegeneration.

Amyloid PET burden classification
Used in Alzheimer's disease trials to confirm amyloid pathology at baseline and measure clearance with anti-amyloid therapies.

Central Vein Sign characterisation
An emerging MRI-based CADx application evaluated for incorporation into updated MS diagnostic criteria — alongside paramagnetic rim lesions — as a more specific imaging biomarker for MS. Research supports its role in improving diagnostic accuracy; prospective validation is ongoing

DAT-SPECT binding ratio quantification
Used in Parkinson's disease trials to measure dopaminergic neuronal integrity and monitor disease progression.

 

Browse QMENTA's full neuroimaging algorithm catalog
Over 50 CADx tools for MS, Alzheimer's, Parkinson's, and oncology — including FreeSurfer, SIENAX, and validated AI biomarker algorithms from leading research institutions.
View the catalog →

 

Computer-aided diagnosis software in clinical trials

Using CAD software in a clinical trial is operationally and regulatorily different from using it in clinical care.

In clinical care, a CAD tool supports the radiologist at the point of care — one input informing a decision about an individual patient. In a clinical trial, CAD outputs may serve as quantitative efficacy endpoints that appear in a regulatory submission. This means the system generating those measurements must meet a higher standard:

  • Analytical validation across scanner types, sites, and time points
  • Version control ensuring every patient's images are processed identically
  • Complete audit trail documenting every processing event
  • Multi-site consistency for the full duration of the study

The version control problem is the most practically significant concern. In a multi-site global trial, site-installed CAD tools create the risk of version drift — different sites running different software versions, producing measurements that are not directly comparable. A cloud-based platform eliminates this: every site's images are processed by the same algorithm, the same version, in the same computational environment.

QMENTA's Imaging Hub provides a catalog of over 50 AI imaging biomarker and CAD analysis tools deployable centrally across multi-site trials.⁶ Algorithm versions are locked at trial initiation and do not drift across sites or across the study duration.

Key takeaways

  • CADe detects; CADx diagnoses, and measures — the distinction matters for trial endpoint use
  • FDA-cleared CAD software supports the clinician; it does not replace clinical judgement
  • QMENTA's Care Platform holds FDA 510(k) clearance K202718 for cloud-based AI neuroimaging
  • In clinical trials, CAD outputs serve as efficacy endpoints and require analytical validation regardless of clearance status
  • Cloud-based CAD deployment solves the version drift problem that affects site-installed tools in multi-site trials
  • Over 50 AI imaging biomarker tools are available in QMENTA's centrally deployed catalog

 

By Paulo Rodrigues, PhD, Chief Technology Officer and Co-Founder at QMENTA
Paulo Rodrigues leads technology strategy at QMENTA and writes about imaging clinical trials, protocol standardization, real-time QC, and compliance-ready neuroimaging workflows for multi-site studies. View executive leadership.

 

¹ FDA. Software as a Medical Device (SaMD). fda.gov/medical-devices/digital-health-center-excellence/software-medical-device-samd

² FDA. Artificial Intelligence and Machine Learning (AI/ML) Software as a Medical Device Action Plan. January 2021. fda.gov

³ FDA 510(k) Premarket Notification Database. K202718. accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm

⁴ Tomas-Fernandez X, Warfield SK. A new classifier feature space for accurate multiple sclerosis lesion segmentation. ISBI 2015.

⁵ Montalban X, Lebrun-Frenay C, Oh J, et al. Diagnosis of multiple sclerosis: 2024 revisions of the McDonald criteria. Lancet Neurol. 2025;24(10):850–865. doi:10.1016/S1474-4422(25)00270-4

⁶ QMENTA. AI Imaging Biomarker Catalog. qmenta.com/imaging-hub/ai-imaging-biomarkers

Explore QMENTA's FDA-cleared CAD platform for neuroimaging

QMENTA's 510(k)-cleared Care Platform (K202718) gives clinicians and clinical researchers access to over 50 AI imaging algorithms for neurological disease — deployable in both care and regulated trial settings.

Explore the platform →

QMENTA-Hub-transparent

Frequently asked questions

What is the difference between CADe and CADx?

CADe — computer-aided detection — identifies and flags potential abnormalities in a medical image for a radiologist to review. It tells the clinician where to look but does not characterise what it has found. CADx — computer-aided diagnosis — goes further, characterising or classifying the abnormality and providing quantitative measurements, probability scores, or severity ratings. CADe is more common in screening workflows; CADx is used in clinical trials and specialised diagnostic applications where quantitative characterisation of a finding is required as an endpoint.

Does FDA-cleared computer-aided diagnosis software make the diagnosis?

No. FDA-cleared computer-aided diagnosis software is classified as a decision support tool — it provides quantitative data and flagged findings for review by a qualified clinician, who remains responsible for the final diagnosis. FDA labelling requirements for cleared CAD software specify that it is intended to assist, not replace, clinical judgement. The clinician reviewing the output retains full professional and legal responsibility for any diagnostic or treatment decision.

Do you need FDA clearance to use computer-aided diagnosis software in a clinical trial?

No. FDA 510(k) clearance is a device marketing authorisation — it governs whether software can be commercially deployed in clinical care. It is not a requirement for using CAD-derived measurements as endpoints in a drug or device clinical trial.

The relevant standard for trial use is analytical validation: the software must demonstrate reproducibility, accuracy, and consistent performance across the scanner types, sites, and patient populations represented in the trial. This requirement applies regardless of clearance status — a cleared tool still needs endpoint-specific validation, and an uncleared tool that is properly validated can legitimately serve as a source of trial endpoint data.

The level of validation required scales with the endpoint's role. Measurements used as primary or secondary efficacy endpoints in a regulatory submission face the highest bar; exploratory endpoints have more flexibility, though the regulatory agency should be consulted on validation expectations during protocol development.

If a CAD-derived measurement is being proposed as a formally recognised novel biomarker, sponsors may pursue qualification through FDA's Drug Development Tool (DDT) programme — but this is a separate track from device clearance and is not required for most trial uses.

In short: clearance matters for clinical deployment; analytical validation is what matters for trial endpoints.

Can computer-aided diagnosis software be used consistently across multiple trial sites?

Yes — and multi-site consistency is one of the primary advantages of cloud-based CAD platforms. A cloud-based system applies the same algorithm version to images from every site, producing measurements that are directly comparable regardless of where the image was acquired. This is not reliably achievable with site-installed software, where version discrepancies between sites can introduce systematic measurement bias. For trials using CAD-derived endpoints, centralised cloud deployment is the preferred architecture.

What neurological conditions can computer-aided diagnosis software analyse?

CAD software for neuroimaging has been developed and in many cases cleared for applications including multiple sclerosis (white matter lesion detection and volume measurement), Alzheimer's disease (brain atrophy quantification, amyloid PET classification), Parkinson's disease (DAT-SPECT binding ratio quantification), traumatic brain injury (axonal and myelin integrity through DTI, haematoma detection and volume measurement), and brain tumours (segmentation and volumetric assessment under RANO criteria). The availability of analytically validated and regulatory-cleared algorithms varies by condition, modality, and region.