Imaging systems and robots overlap through sensing, positioning and control. A scanner is not automatically a robot. The following are introductory engineering explanations and proposed learning tests, not manufacturer build instructions.
MRI · Magnetic resonance
Construction & architecture
Main magnet, gradient coils, radiofrequency coils and receiver electronics.
How it operates
Magnet → RF excitation → spatial encoding → received signal → reconstructed slices
Conceptual signal path; component arrangements vary by manufacturer.Test & challenge
Check field uniformity, geometric accuracy and phantom image quality; assess magnetic, heating and acoustic hazards.
Learning exercise: identify a likely artefact, propose a controlled test, then explain what evidence would support a correction.
A condition-focused layer: research signals, practical support and accessible design.
Terms & scope
ASD and autism refer to the same spectrum. Asperger’s is a historical diagnostic term within current autism descriptions; preserve it when discussing older diagnoses or papers. NHS Dorset · terminology
Autism and ADHD are neurodevelopmental conditions. Anxiety, depression and other mental-health needs may coexist and require their own evidence and support. NHS · autism and co-occurrence · NHS · adult ADHD
Accessibility & safeguarding layer
Proposed design checklist: predictable instructions, adjustable sensory demands, breaks, clear consent and an accessible way to stop. For children, include appropriate consent and assent. Ask each person what works for them; do not assume one autistic profile.
Co-design goals around communication, independence and chosen participation. Record distress, adverse effects and privacy risks alongside performance. Avoid optimising apparent compliance or masking as a substitute for wellbeing.
Research, diagnosis & support
Keep three questions separate: what is measured, what is inferred, and whether using the technology improves a person’s life. NICE advises against routine neuroimaging for adult autism diagnosis. NICE CG142 · clinical assessment
Findings catalogue
Selected starting catalogue, not a systematic or exhaustive review. Abstract-based records are marked through their limits. Evidence categories describe study design, not a formal certainty rating. Published findings belong to the named authors; these summaries and engineering questions are editorial synthesis. Reviewed 10 October 2026.
12 records
MRI feature fusion · Dekhil et al. Retrospective classification
Autism · Mixed / unspecified · 2020
Finding: Across 18 ABIDE datasets (1,060 participants), the abstract reports accuracy ranges of 75–100% for structural MRI and 79–100% for fMRI.
Limit: Ranges across datasets are not one clinical accuracy estimate. Abstract-level catalogue entry; age-specific performance and independent clinical validation are not established here.
Engineering question: Check participant-level splits, site effects, calibration, missing modalities and external validation before claiming generalisation.
Adult structural connectivity · diffusion MRI Observational research
ADHD · Adults · 2023
Finding: Researchers compared diffusion-derived connectivity in 40 adults with ADHD and 36 controls.
Limit: This entry records the study design, not a verified diagnostic biomarker. The adult sample must not be treated as an exclusively over-25 group.
Engineering question: Ask how tractography settings and head motion affect the estimated networks; verify full results before recording a direction of effect.
Graph attention classification · Cai et al. Small-sample AI experiment
Autism · Children / young people · 2025
Finding: A graph attention model using resting fNIRS data from 25 autistic and 22 typically developing children reported 97.92% accuracy.
Limit: This is the authors’ experimental result, not clinical diagnostic performance. Full split design, confidence intervals and external replication need appraisal.
Engineering question: Audit participant separation, window leakage, class balance, motion and optical coupling; test an independent cohort.
Neurofeedback with an active comparison · Barth et al. Randomised trial · no superiority
ADHD · Adults · 2021
Finding: SCP neurofeedback, fNIRS neurofeedback and EMG biofeedback groups all improved; neither neurofeedback approach showed superior symptom effects over the comparison.
Limit: Adult eligibility began at 18; no exclusively over-25 finding is claimed. Improvement within a group does not prove a technology-specific effect.
Engineering question: Retain credible controls, blind outcome assessment where possible and track learning separately from symptom ratings.
Robot-assisted PRT · children aged 3–8 Exploratory randomised trial
Autism · Children / young people · 2021 / 2022
Finding: Among 73 assessed children, groups did not differ significantly on reported general social-communication/global-function outcomes directly after treatment; the largest follow-up gains occurred with robot-assisted PRT.
Limit: The robot was part of a supported intervention, not an autonomous cure. Delayed results and different outcomes need separate interpretation.
Engineering question: Separate robot contribution from therapist contact, novelty and treatment content; assess transfer beyond the robot.
VR-supported phobia intervention · adult study Small pilot + individual account
Autism + anxiety · Adults · over-25 case · 2019
Finding: A pilot involved eight autistic adults aged 18–57. The university reports maintained everyday phobia-related improvement in five at six months; its individual account features Rebecca, aged 26.
Limit: The cohort was mixed-age, not all over 25. An uncontrolled pilot and personal account do not establish efficacy for every adult or isolate VR from CBT.
Engineering question: Separate the immersive display from clinician-guided graded exposure; measure real-world goals and longer-term outcomes.
QbTest as an assessment adjunct Clinical guidance · bounded use
ADHD · Children / young people · 2024 guidance
Finding: NICE recommends QbTest as an option to help diagnose ADHD at ages 6–17 alongside specialist assessment.
Limit: This is not a scan or standalone diagnostic AI. Do not extend this child recommendation to adult diagnosis; NICE identifies adult evidence gaps.
Engineering question: Evaluate workflow benefit, accessibility and differential assessment, not just test-score separation.
Finding: No autism- or ADHD-specific outcome study has been appraised for this technology in this selected catalogue.
Limit: This means a gap in this catalogue, not proof that research does not exist. General imaging capability must not be converted into an autism benefit claim.
Engineering question: Next review: identify a primary study, confirm population and purpose, then record findings, safety and limitations.
PET / SPECT · autism / ADHD evidence queue No condition-specific finding catalogued
Evidence gap · Not established · Review pending
Finding: No autism- or ADHD-specific outcome study has been appraised for this technology in this selected catalogue.
Limit: This means a gap in this catalogue, not proof that research does not exist. General imaging capability must not be converted into an autism benefit claim.
Engineering question: Next review: identify a primary study, confirm population and purpose, then record findings, safety and limitations.
Finding: No autism- or ADHD-specific outcome study has been appraised for this technology in this selected catalogue.
Limit: This means a gap in this catalogue, not proof that research does not exist. General imaging capability must not be converted into an autism benefit claim.
Engineering question: Next review: identify a primary study, confirm population and purpose, then record findings, safety and limitations.
Record authors, date, DOI/URL, population and ages, sample size, technology, comparator, outcome and follow-up. Then record uncertainty, adverse events, funding/conflicts and replication. Mark absent information as unverified.
For AI: check leakage, external validation, subgroup performance and calibration. For hardware: check signal quality and measurement repeatability. For intervention: check benefit beyond practice, novelty or clinician contact. A high classifier score alone is not a proven support outcome.
Case studies · separate age groups
Children · robot-assisted communication support
Study: van den Berk-Smeekens and colleagues, online 2021 / journal 2022. Seventy-three assessed autistic children aged 3–8 received PRT, robot-assisted PRT or usual care.
Outcome: No significant group differences on the abstract’s reported general outcomes immediately after treatment; largest gains at follow-up occurred in the robot-assisted group.
Takeaway: Test generalisation and delayed outcomes. This is an exploratory trial, not proof that robots replace human support. Primary trial · methods and results
Adults over 25 · Rebecca, aged 26 in the published account
Context: Newcastle University’s 2019 account describes an autistic woman whose doorway/corridor phobia interrupted university attendance. It reports four Blue Room sessions and subsequent practice in everyday settings.
Technology: Personalised room-scale VR with therapist-guided exposure; no headset was required. VR itself is not necessarily AI.
Reported outcome: She described greater independence and returning to university, completing her degree. Her age is 26 at publication; age at treatment is not established here.
Takeaway: Judge transfer to personally meaningful goals. This individual account supports understanding lived outcomes, not a controlled estimate of benefit for all adults over 25.
Different systems follow different evidence routes. Engineering approval, market conformity and hospital acceptance are distinct decisions.
1 · Define purpose
Specify users, use environment, measurement and acceptance criteria.
Responsibility & evidence: Systems engineer and clinical/research partners: agreed requirements and intended use.
2 · Design architecture
Allocate mechanical, electrical, software and safety responsibilities.
Responsibility & evidence: Design team: drawings, interfaces, signal budget and traceable risk controls.
3 · Build prototype
Integrate components and document configurations and calibration.
Responsibility & evidence: Engineering team: controlled build records and bench results.
4 · Verify requirements
Test each specification against predefined acceptance criteria.
Responsibility & evidence: Test and quality teams: protocols, results, deviations and requirement traceability.
5 · Validate use
Evaluate whether the complete system meets intended user needs.
Responsibility & evidence: Relevant users and clinical specialists: usability and performance evidence appropriate to the device.
6 · Review conformity
Determine the applicable market, device classification and conformity route.
Responsibility & evidence: Manufacturer and, where required, assessment body: technical documentation and conformity evidence. Routes differ by jurisdiction and class.
7 · Accept installation
Check site readiness, installation, commissioning and user training.
Responsibility & evidence: Local service and clinical teams: acceptance records and operational readiness. This is separate from market conformity.
8 · Monitor operation
Maintain calibration, service records, incident handling and change control.
Responsibility & evidence: Operator and manufacturer: monitoring evidence; changes may require renewed testing and regulatory review.
AI may support reconstruction, segmentation, artefact detection or robotic perception. These are possible application areas, not capabilities of every system. An agent may coordinate bounded software tasks; authority to affect physical equipment requires explicit limits, safety controls and validated behaviour.
Safeguarding & cybersecurity
Proposed review prompts: who may command movement or acquisition, what happens on lost communication, what is logged, how patient data is protected, and how updates are tested and rolled back? Separate safety controls from model confidence; uncertain predictions must not silently authorise hazardous actions.
Learn · test · explain
Use synthetic signals or public demonstration data to introduce noise, timing errors or missing measurements. Compare a baseline with a correction, record assumptions and explain residual uncertainty. Physical fault testing belongs in an authorised controlled facility.
Self-check: Which component senses? Which component acts? What is measured directly? What is inferred? Who accepts the evidence at each gate?
Research directions and exercises proposed for learning; no hardware experiment or qualification is claimed. Reviewed 10 October 2026.