AI implementation in the UK landscape: Knowledge of AI governance, perceived challenges and opportunities, and ways forward for radiographers

INTRODUCTION

Despite the rapid increase of AI-enabled applications deployed in clinical practice, many challenges exist around AI implementation, including the clarity of governance frameworks, usability of validation of AI models, and customisation of training for radiographers. This study aimed to explore the perceptions of diagnostic and therapeutic radiographers, with existing theoretical and/or practical knowledge of AI, on issues of relevance to the field, such as AI implementation, including knowledge of AI governance and procurement, perceptions about enablers and challenges and future priorities for AI adoption.

METHODS

An online survey was designed and distributed to UK-based qualified radiographers who work in medical imaging and/or radiotherapy and have some previous theoretical and/or practical knowledge of working with AI. Participants were recruited through the researchers' professional networks on social media with support from the AI advisory group of the Society and College of Radiographers. Survey questions related to AI training/education, knowledge of AI governance frameworks, data privacy procedures, AI implementation considerations, and priorities for AI adoption. Descriptive statistics were employed to analyse the data, and chi-square tests were used to explore significant relationships between variables.

RESULTS

In total, 88 valid responses were received. Most radiographers (56.6 %) had not received any AI-related training. Also, although approximately 63 % of them used an evaluation framework to assess AI models' performance before implementation, many (36.9 %) were still unsure about suitable evaluation methods. Radiographers requested clearer guidance on AI governance, ample time to implement AI in their practice safely, adequate funding, effective leadership, and targeted support from AI champions. AI training, robust governance frameworks, and patient and public involvement were seen as priorities for the successful implementation of AI by radiographers.

CONCLUSION

AI implementation is progressing within radiography, but without customised training, clearer governance, key stakeholder engagement and suitable new roles created, it will be hard to harness its benefits and minimise related risks.

IMPLICATIONS FOR PRACTICE

The results of this study highlight some of the priorities and challenges for radiographers in relation to AI adoption, namely the need for developing robust AI governance frameworks and providing optimal AI training.

Assessing Bladder Radiotherapy Response With Quantitative Diffusion-Weighted Magnetic Resonance Imaging Analysis

AIMS

Radiotherapy with radiosensitisation offers opportunity for cure with organ preservation in muscle-invasive bladder cancer (MIBC). Treatment response assessment and follow-up are reliant on regular endoscopic evaluation of the retained bladder. In this study we aim to determine the role of diffusion-weighted magnetic resonance imaging (DWI) and apparent diffusion coefficient (ADC) analysis to assess bladder radiotherapy response.

MATERIALS AND METHODS

Patients with T2-T4aN0-3M0 MIBC suitable for radical radiotherapy were recruited prospectively to an ethics approved protocol. Following transurethral resection of the bladder tumour and prior to any treatment, magnetic resonance imaging including DWI was performed on a 1.5T system using b values of 0, 100, 150, 250, 500, 750 s/mm2. DWI was repeated 3 months after completing radiotherapy. Cystoscopy and tumour site biopsy were undertaken following this. The response was dichotomised into response (

RESULTS

Thirty-four patients were evaluated. Response was associated with a significant increase in ΔADC mean compared with poor response at ΔADCall (0.57 × 10-3 mm2/s versus -0.01 × 10-3 mm2/s; P < 0.0001) and ΔADCb100 (0.58 × 10-3 mm2/s versus -0.10 x 10-3 mm2/s; P = 0.007). A 48.50% increase in %ΔADCall mean was seen in response compared with a 1.37% decrease in poor response (P < 0.0001). This corresponded to a %ΔADCb100 mean increase of 50.34% in response versus a 7.36% decrease for poor response (P < 0.0001). Significant area under the curve (AUC) values predictive of radiotherapy response were identified at ΔADC and %ΔADC for ADCall and ADCb100 mean, 10th, 25th, 50th, 75th and 90th percentiles (AUC >0.9, P < 0.01). ΔADCall mean of 0.16 × 10-3 mm2/s and ΔADCb100 mean 0.12 × 10-3 mm2/s predicted radiotherapy response with sensitivity/specificity/positive predictive value/negative predictive value of 92.9%/100.0%/100.0%/75.0% and 89.3%/100.0%/100.0%/66.7%, respectively.

CONCLUSIONS

Quantitative DWI analysis can successfully provide non-invasive assessment of bladder radiotherapy response. Multicentre validation is required before prospective testing to inform MIBC radiotherapy follow-up schedules and decision making.

Collapse

Key Words

• Bladder cancer
• Diffusion-weighted MRI
• MRI
• Radiotherapy

Collapse

MESH Headings

• Diffusion Magnetic Resonance Imaging/methods
• Humans
• Magnetic Resonance Imaging
• Prospective Studies
• ROC Curve
• Urinary Bladder/diagnostic imaging
• Urinary Bladder Neoplasms/diagnostic imaging
• Urinary Bladder Neoplasms/pathology
• Urinary Bladder Neoplasms/radiotherapy

Collapse

Grants

• 28724 Cancer Research UK
• C33589/A19727 Cancer Research UK
• Department of Health
• C33589/A28284 Cancer Research UK

Collapse

Affiliation(s)

S Hafeez The Institute of Cancer Research, London, UK; The Royal Marsden NHS Foundation Trust, London, UK.
M Koh The Institute of Cancer Research, London, UK; The Royal Marsden NHS Foundation Trust, London, UK
K Jones The Institute of Cancer Research, London, UK; The Royal Marsden NHS Foundation Trust, London, UK
A El Ghzal The Institute of Cancer Research, London, UK; The Royal Marsden NHS Foundation Trust, London, UK
J D'Arcy The Institute of Cancer Research, London, UK; The Royal Marsden NHS Foundation Trust, London, UK
P Kumar The Royal Marsden NHS Foundation Trust, London, UK
V Khoo The Royal Marsden NHS Foundation Trust, London, UK
S Lalondrelle The Institute of Cancer Research, London, UK; The Royal Marsden NHS Foundation Trust, London, UK
F McDonald The Institute of Cancer Research, London, UK; The Royal Marsden NHS Foundation Trust, London, UK
A Thompson The Royal Marsden NHS Foundation Trust, London, UK
E Scurr The Royal Marsden NHS Foundation Trust, London, UK
A Sohaib The Royal Marsden NHS Foundation Trust, London, UK
R Huddart The Institute of Cancer Research, London, UK; The Royal Marsden NHS Foundation Trust, London, UK

Collapse

Patient deprivation and perceived scan burden negatively impact the quality of whole-body MRI

AIM

To evaluate the association between the image quality of cancer staging whole-body magnetic resonance imaging (WB-MRI) and patient demographics, distress, and perceived scan burden.

MATERIALS AND METHODS

A sample of patients recruited prospectively to multicentre trials comparing WB-MRI with standard scans for staging lung and colorectal cancer were invited to complete two questionnaires. The baseline questionnaire, administered at recruitment, collated data on demographics, distress and co-morbidity. The follow-up questionnaire, completed after staging investigations, measured perceived WB-MRI scan burden (scored 1 low to 7 high). WB-MRI anatomical coverage, and technical quality was graded by a radiographic technician and grading combined to categorise the scan as "optimal", "sub-optimal" or "degraded". A radiologist categorised 30 scans to test interobserver agreement. Data were analysed using the chi-square, Fisher's exact, t-tests, and multinomial regression.

RESULTS

One hundred and fourteen patients were included in the study (53 lung, 61 colorectal; average age 65.3 years, SD=11.8; 66 men [57.9%]). Overall, 45.6% (n=52), scans were classified as "optimal" quality, 39.5% (n=45) "sub-optimal", and 14.9% (n=17) as "degraded". In adjusted analyses, greater deprivation level and higher patient-reported scan burden were both associated with a higher likelihood of having a sub-optimal versus an optimal scan (odds ratio [OR]: 4.465, 95% confidence interval [CI]: 1.454 to 13.709, p=0.009; OR: 1.987, CI: 1.153 to 3.425, p=0.013, respectively). None of the variables predicted the likelihood of having a degraded scan.

CONCLUSIONS

Deprivation and patients' perceived experience of the WB-MRI are related to image quality. Tailored protocols and individualised patient management before and during WB-MRI may improve image quality.

Blurring the lines for better visualisation

On-treatment imaging in radiotherapy has evolved over the last 60 years, bringing with it changes in the roles of radiographers, radiologists and oncologists. The ability to acquire and interpret high quality images (2D kilovoltage and megavoltage imaging and 3D CT and cone-beam CT) for radiotherapy planning and delivery requires therapy radiographers to have skills and knowledge that overlap with those of diagnostic radiographers. With the implementation of MRI-guided radiotherapy, treatment radiographers and clinical oncologists are exploring new territory, requiring truly collaborative working practices with their radiology partners. This short communication introduces the first images acquired using the hybrid MR Linac at our institution.

First MRI application of an active breathing coordinator

A commercial active breathing coordinator (ABC) device, employed to hold respiration at a specific level for a predefined duration, was successfully adapted for magnetic resonance imaging (MRI) use for the first time. Potential effects of the necessary modifications were assessed and taken into account. Automatic MR acquisition during ABC breath holding was achieved. The feasibility of MR-ABC thoracic and abdominal examinations together with the advantages of imaging in repeated ABC-controlled breath holds were demonstrated on healthy volunteers. Five lung cancer patients were imaged under MR-ABC, visually confirming the very good intra-session reproducibility of organ position in images acquired with the same patient positioning as used for computed tomography (CT). Using identical ABC settings, good MR-CT inter-modality registration was achieved. This demonstrates the value of ABC, since application of T1, T2 and diffusion weighted MR sequences provides a wider range of contrast mechanisms and additional diagnostic information compared to CT, thus improving radiotherapy treatment planning and assessment.

Breast dynamic contrast-enhanced examinations with fat suppression: are contrast-agent uptake curves affected by magnetic field inhomogeneity?

OBJECTIVES

To investigate the effect of magnetic field heterogeneity in breast dynamic contrast-enhanced examinations with fat saturation (DCE-FS).

METHODS

The magnetic field was mapped over the breasts in ten patients. DCE-FS was undertaken at 1.5 T with fast spoiled gradient echoes and spectrally selective fat saturation. Signal intensity was calculated for T1 values 25-1,200 ms both on and off resonance, and results were verified with a test object. Clinical examinations were evaluated for the predicted effects of field heterogeneity.

RESULTS

Magnetic field was found to vary by 3.6 ± 1.2 ppm over the central transaxial slice and 5.1 ± 1.5 over the whole breast volume (mean ± standard deviation). Computer simulations predict a reduction in the dynamic range if field heterogeneity leads to unintended water suppression, and distortion to CA uptake curves due to fat suppression failure (for fat containing pixels). A compromise between dynamic range and fat saturation performance is required. Both water suppression and fat suppression failure are apparent in clinical examinations.

CONCLUSION

Magnetic field heterogeneity is likely to reduce the sensitivity of DCE-FS by distorting the CA uptake curves because of fat suppression failure (for fat containing pixels) and by reducing the dynamic range because of unintended water suppression.

KEY POINTS

• Magnetic field heterogeneity is significant in breast magnetic resonance. • Contrast-agent uptake curves are distorted by a non-uniform magnetic field. • Radiologist must be aware of possibility of distortion to interpret uptake curves correctly. • Compromise between fat suppression and dynamic range is required.

[Leprosy, a pillar of human genetics of infectious diseases]

Despite a natural reservoir of Mycobacterium leprae limited to humans and free availability of an effective antibiotic treatment, more than 200,000 people develop leprosy each year. This disease remains a major cause of disability and social stigma worldwide. The cause of this constant incidence is currently unknown and indicates that important aspects of the complex relationship between the pathogen and its human host remain to be discovered. An important contribution of host genetics to susceptibility to leprosy has long been suggested to account for the considerable variability between individuals sustainably exposed to M. leprae. Given the inability to cultivate M. leprae in vitro and in the absence of relevant animal model, genetic epidemiology is the main strategy used to identify the genes and, consequently, the immunological pathways involved in protective immunity to M. leprae. Recent genome-wide studies have identified new pathophysiological pathways which importance is only beginning to be understood. In addition, the prism of human genetics placed leprosy at the crossroads of other common diseases such as Crohn's disease, asthma or myocardial infarction. Therefore, novel lights on the pathogenesis of many common diseases could eventually emerge from the detailed understanding of a disease of the shadows.

Assessment of colorectal hepatic metastases by quantitative T2 relaxation time

AIM

To determine the T(2) relaxation time of colorectal hepatic metastases and changes in T(2) relaxation times following chemotherapy.

MATERIALS AND METHODS

42 patients with 96 hepatic colorectal metastases underwent baseline MRI. Axial T(1), T(2) and multi-echo GRASE sequences were acquired. ROIs were drawn on T(2) relaxation maps, obtained from GRASE images, encompassing metastasis and normal liver to record T(2) relaxation time values. In 11 patients with 28 metastases, MRI was repeated using same protocol at 6 weeks following chemotherapy. The median pre-treatment T(2) values of metastases and normal liver were compared using the Mann-Whitney test. The pre- and post-treatment median T(2) values of metastases were compared using the Wilcoxon-Rank test for responding (n=16) and non-responding (n=12) lesions defined by RECIST criteria. The change in T(2) values (ΔT(2)) were compared and correlated with percentage change in lesion size.

RESULTS

There was no difference in the pre-treatment median T(2) of metastases between responding (67.3±8.6) and non-responding metastases (71.4±16.5). At the end of chemotherapy, there was a decrease in the median T(2) of responding lesions (61.6±12.6) p=0.83, and increase in non-responding lesions (76.2±18.4) p=0.03, but these were not significantly different from the pre-treatment values. There was no significant difference in ΔT(2) of responding and non-responding lesions (p=0.18) and no correlation was seen between size change and ΔT(2) (coefficient=0.3).

CONCLUSION

T(2) relaxation time does not appear to predict response of colorectal liver metastasis to chemotherapy.

Optimising magnetic resonance imaging for preoperative staging and surgical planning in colon cancer at 1.5 tesla and 3.0 tesla

395

Background: The aim of this study was to determine best technique for future preoperative staging and operative planning in colon cancer using MRI at 1.5T, and feasibility at 3.0T.

Methods: 39 scans were performed in 4 healthy volunteers comparing bowel distension using different oral contrast techniques; no oral preparation, 1L of water, 1L of poly ethylene glycol (Klean-Prep), and 1L of 2.1% barium sulphate and 3.2% sorbitol solution (E-Z-Cat) at 1.5T. Bowel distension was scored blindly by 2 experienced radiologists. A further 35 scans compared image quality of T2-weighted pulse sequences; 2D-TrueFISP, 3D-TrueFISP, HASTE, 2D-TSE and 3D-TSE. Scans were scored for movement artefact, colon wall and colon mesentery visualisation. Signal to noise ratio (SNR) was calculated. Inter-observer agreement was assessed using Cohen's Kappa. Equivalent optimised protocols were derived at 3.0T and tested for feasibility.

Results: Inter-observer agreement for scoring was good (kappa = 0.61-0.67). Sorbitol-barium sulphate gave significantly better small bowel distension than no oral preparation, water and Klean-Prep (mean score 3.8 vs 1.6, 2.2 and 1.9, p≤0.007). 2D TrueFISP sequences gave the highest mean score for scan quality, significantly better than 3D-TSE, 2D-TSE, 3D-TrueFISP and HASTE (mean 10.6 vs 3.2, 7.8, 7.7 and 7.4, p≤0.012). 2D-TSE sequences obtained best SNR, significantly greater than obtained using 3D-TSE (p=0.045). 3.0T sequences were comparable or superior.

Conclusions: Colon imaging using MRI requires fast sequences and adequate bowel distension. Our observations indicate that 2D-TrueFISP and 2D-TSE sequences with 1L Sorbitol-Barium sulphate orally give the best T2-weighted images of the colon wall and adjacent structures for cancer staging at 1.5T; equivalent 3.0T protocols are possible.

No significant financial relationships to disclose.

The development and optimization of high spatial resolution MRI for imaging the oesophagus using an external surface coil

This paper describes the development and optimization of an innovative technique using an external surface coil to obtain high resolution, thin section MR images of the oesophagus using volunteers. T2 weighted fast spin echo sequences were performed with and without cardiac gating. The field of view (FOV), matrix size, slice thickness, number of signal averages (NSA), and repetition time (TR)/echo time (TE) were altered to optimize signal to noise ratio (SNR) whilst maintaining spatial resolution. The effect of cardiac gating was also investigated. Workstation images were evaluated on the ability to visualize: individual oesophageal wall layers; perioesophageal fat; the azygos vein and wall of the descending aorta, giving qualitative assessment of image clarity. The optimum sequence enabled the layers of the oesophageal wall and perioesophageal tissues to be demonstrated in an acceptable scan time of 7.07 min. A FOV of less than 250 mm degraded image quality so that individual oesophageal wall layers could not be depicted and noise within the image impaired visualization of posterior mediastinal structures. The results indicate that high resolution imaging of the oesophagus using an external surface coil can depict anatomic structures clearly and that the use of cardiac gating improves image clarity. The technique offers an alternative, non-invasive method of detailed imaging of the oesophagus.

Colorectal hepatic metastases: quantitative measurements using single-shot echo-planar diffusion-weighted MR imaging

The purpose of this study was to obtain quantitative measurements of the apparent diffusion coefficient (ADC1), flow insensitive apparent diffusion coefficient (ADC2) and perfusion fraction (F) of colorectal hepatic metastases using DWI and to compare these measurements with those obtained in liver parenchyma. Forty patients with 66 hepatic metastases from colorectal carcinoma were prospectively evaluated using DWI with three b values. Quantitative maps of the ADC1 (using b = 0, 150, 500 s/mm2 images), ADC2 (using b = 150, 500 s/mm2 images) and fractional variation (F) between ADC1 and ADC2, which reflects perfusion fraction, were calculated. The ADC1, ADC2 and F derived from metastases and liver parenchyma were compared. The mean ADC1 values of liver parenchyma and metastases were significantly higher than the mean ADC2 values (P < 0.0001, paired t-test). Colorectal metastases were found to have higher mean ADC1 and ADC2 values compared with liver (P < 0.0001, Mann-Whitney test). However, the estimated F was found to be lower in metastases compared to liver (P = 0.03, Mann-Whitney test). Colorectal hepatic metastases were characterised by higher ADC1 and ADC2 values, but lower F values compared to liver.