Black box no more: A cross-sectional multi-disciplinary survey for exploring governance and guiding adoption of AI in medical imaging and radiotherapy in the UK.

BACKGROUND: Medical Imaging and radiotherapy (MIRT) are at the forefront of artificial intelligence applications. The exponential increase of these applications has made governance frameworks necessary to uphold safe and effective clinical adoption. There is little information about how healthcare practitioners in MIRT in the UK use AI tools, their governance and associated challenges, opportunities and priorities for the future. METHODS: This cross-sectional survey was open from November to December 2022 to MIRT professionals who had knowledge or made use of AI tools, as an attempt to map out current policy and practice and to identify future needs. The survey was electronically distributed to the participants. Statistical analysis included descriptive statistics and inferential statistics on the SPSS statistical software. Content analysis was employed for the open-ended questions. RESULTS: Among the 245 responses, the following were emphasised as central to AI adoption: governance frameworks, practitioner training, leadership, and teamwork within the AI ecosystem. Prior training was strongly correlated with increased knowledge about AI tools and frameworks. However, knowledge of related frameworks remained low, with different professionals showing different affinity to certain frameworks related to their respective roles. Common challenges and opportunities of AI adoption were also highlighted, with recommendations for future practice.

Black box no more: a scoping review of AI governance frameworks to guide procurement and adoption of AI in medical imaging and radiotherapy in the UK.

Technological advancements in computer science have started to bring artificial intelligence (AI) from the bench closer to the bedside. While there is still lots to do and improve, AI models in medical imaging and radiotherapy are rapidly being developed and increasingly deployed in clinical practice. At the same time, AI governance frameworks are still under development. Clinical practitioners involved with procuring, deploying, and adopting AI tools in the UK should be well-informed about these AI governance frameworks. This scoping review aimed to map out available literature on AI governance in the UK, focusing on medical imaging and radiotherapy. Searches were performed on Google Scholar, Pubmed, and the Cochrane Library, between June and July 2022. Of 4225 initially identified sources, 35 were finally included in this review. A comprehensive conceptual AI governance framework was proposed, guided by the need for rigorous AI validation and evaluation procedures, the accreditation rules and standards, and the fundamental ethical principles of AI. Fairness, transparency, trustworthiness, and explainability should be drivers of all AI models deployed in clinical practice. Appropriate staff education is also mandatory to ensure AI's safe and responsible use. Multidisciplinary teams under robust leadership will facilitate AI adoption, and it is crucial to involve patients, the public, and practitioners in decision-making. Collaborative research should be encouraged to enhance and promote innovation, while caution should be paid to the ongoing auditing of AI tools to ensure safety and clinical effectiveness.

Test-retest repeatability of ADC in prostate using the multi b-Value VERDICT acquisition.

PURPOSE: VERDICT (Vascular, Extracellular, Restricted Diffusion for Cytometry in Tumours) MRI is a multi b-value, variable diffusion time DWI sequence that allows generation of ADC maps from different b-value and diffusion time combinations. The aim was to assess precision of prostate ADC measurements from varying b-value combinations using VERDICT and determine which protocol provides the most repeatable ADC. MATERIALS AND METHODS: Forty-one men (median age: 67.7 years) from a prior prospective VERDICT study (April 2016-October 2017) were analysed retrospectively. Men who were suspected of prostate cancer and scanned twice using VERDICT were included. ADC maps were formed using 5b-value combinations and the within-subject standard deviations (wSD) were calculated per ADC map. Three anatomical locations were analysed per subject: normal TZ (transition zone), normal PZ (peripheral zone), and index lesions. Repeated measures ANOVAs showed which b-value range had the lowest wSD, Spearman correlation and generalized linear model regression analysis determined whether wSD was related to ADC magnitude and ROI size. RESULTS: The mean lesion ADC for b0b1500 had the lowest wSD in most zones (0.18-0.58x10-4 mm2/s). The wSD was unaffected by ADC magnitude (Lesion: p = 0.064, TZ: p = 0.368, PZ: p = 0.072) and lesion Likert score (p = 0.95). wSD showed a decrease with ROI size pooled over zones (p = 0.019, adjusted regression coefficient = -1.6x10-3, larger ROIs for TZ versus PZ versus lesions). ADC maps formed with a maximum b-value of 500 s/mm2 had the largest wSDs (1.90-10.24x10-4 mm2/s). CONCLUSION: ADC maps generated from b0b1500 have better repeatability in normal TZ, normal PZ, and index lesions.

Incidental findings on brain imaging and blood tests: results from the first phase of Insight 46, a prospective observational substudy of the 1946 British birth cohort.

OBJECTIVE: To summarise the incidental findings detected on brain imaging and blood tests during the first wave of data collection for the Insight 46 study. DESIGN: Prospective observational sub-study of a birth cohort. SETTING: Single-day assessment at a research centre in London, UK. PARTICIPANTS: 502 individuals were recruited from the MRC National Survey of Health and Development (NSHD), the 1946 British birth cohort, based on pre-specified eligibility criteria; mean age was 70.7 (SD: 0.7) and 49% were female. OUTCOME MEASURES: Data regarding the number and types of incidental findings were summarised as counts and percentages, and 95% confidence intervals were calculated. RESULTS: 93.8% of participants completed a brain scan (n=471); 4.5% of scanned participants had a pre-defined reportable abnormality on brain MRI (n=21); suspected vascular malformations and suspected intracranial mass lesions were present in 1.9% (n=9) and 1.5% (n=7) respectively; suspected cerebral aneurysms were the single most common vascular abnormality, affecting 1.1% of participants (n=5), and suspected meningiomas were the most common intracranial lesion, affecting 0.6% of participants (n=3); 34.6% of participants had at least one abnormality on clinical blood tests (n=169), but few reached the prespecified threshold for urgent action (n=11). CONCLUSIONS: In older adults, aged 69-71 years, potentially serious brain MRI findings were detected in around 5% of participants, and clinical blood test abnormalities were present in around one third of participants. Knowledge of the expected prevalence of incidental findings in the general population at this age is useful in both research and clinical settings.

First-in-human in vivo non-invasive assessment of intra-tumoral metabolic heterogeneity in renal cell carcinoma.

Intratumoral genetic heterogeneity and the role of metabolic reprogramming in renal cell carcinoma (RCC) have been extensively documented. However, the distribution of these metabolic changes within the tissue has not been explored. We report on the first-in-human in vivo non-invasive metabolic interrogation of RCC using hyperpolarized carbon-13 (13C) magnetic resonance imaging (HP-MRI) and describe the validation of in vivo lactate metabolic heterogeneity against multi-regional ex vivo mass spectrometry. HP-MRI provides an in vivo assessment of metabolism and provides a novel opportunity to safely and non-invasively assess cancer heterogeneity.

Filtration-histogram based magnetic resonance texture analysis (MRTA) for glioma IDH and 1p19q genotyping.

BACKGROUND: To determine if filtration-histogram based texture analysis (MRTA) of clinical MR imaging can non-invasively identify molecular subtypes of untreated gliomas. METHODS: Post Gadolinium T1-weighted (T1+Gad) images, T2-weighted (T2) images and apparent diffusion coefficient (ADC) maps of 97 gliomas (54 = WHO II, 20 = WHO III, 23 = WHO IV) between 2010 and 2016 were studied. Whole-tumor segmentations were performed on a proprietary texture analysis research platform (TexRAD, Cambridge, UK) using the software's freehand drawing tool. MRTA commences with a filtration step, followed by quantification of texture using histogram texture parameters. Results were correlated using non-parametric statistics with a logistic regression model generated. RESULTS: T1+Gad performed best for IDH typing of glioblastoma (sensitivity 91.9%, specificity 100%, AUC 0.945) and ADC for non-Gadolinium-enhancing gliomas (sensitivity 85.7%, specificity 78.4%, AUC 0.877). T2 was moderately precise (sensitivity 83.1%, specificity 78.9%, AUC 0.821). Excellent results for IDH typing were achieved from a combination of the three sequences (sensitivity 90.5%, specificity 94.5%, AUC = 0.98). For discriminating 1p19q genotypes, ADC produced the best results using unfiltered textures (sensitivity 80.6%, specificity 89.3%, AUC 0.811). CONCLUSION: Preoperative glioma genotyping with MRTA appears valuable with potential for clinical translation. The optimal choice of texture parameters is influenced by sequence choice, tumour morphology and segmentation method.

UK quantitative WB-DWI technical workgroup: consensus meeting recommendations on optimisation, quality control, processing and analysis of quantitative whole-body diffusion-weighted imaging for cancer.

OBJECTIVE: Application of whole body diffusion-weighted MRI (WB-DWI) for oncology are rapidly increasing within both research and routine clinical domains. However, WB-DWI as a quantitative imaging biomarker (QIB) has significantly slower adoption. To date, challenges relating to accuracy and reproducibility, essential criteria for a good QIB, have limited widespread clinical translation. In recognition, a UK workgroup was established in 2016 to provide technical consensus guidelines (to maximise accuracy and reproducibility of WB-MRI QIBs) and accelerate the clinical translation of quantitative WB-DWI applications for oncology. METHODS: A panel of experts convened from cancer centres around the UK with subspecialty expertise in quantitative imaging and/or the use of WB-MRI with DWI. A formal consensus method was used to obtain consensus agreement regarding best practice. Questions were asked about the appropriateness or otherwise on scanner hardware and software, sequence optimisation, acquisition protocols, reporting, and ongoing quality control programs to monitor precision and accuracy and agreement on quality control. RESULTS: The consensus panel was able to reach consensus on 73% (255/351) items and based on consensus areas made recommendations to maximise accuracy and reproducibly of quantitative WB-DWI studies performed at 1.5T. The panel were unable to reach consensus on the majority of items related to quantitative WB-DWI performed at 3T. CONCLUSION: This UK Quantitative WB-DWI Technical Workgroup consensus provides guidance on maximising accuracy and reproducibly of quantitative WB-DWI for oncology. The consensus guidance can be used by researchers and clinicians to harmonise WB-DWI protocols which will accelerate clinical translation of WB-DWI-derived QIBs.

Joint PET-MR respiratory motion models for clinical PET motion correction.

Patient motion due to respiration can lead to artefacts and blurring in positron emission tomography (PET) images, in addition to quantification errors. The integration of PET with magnetic resonance (MR) imaging in PET-MR scanners provides complementary clinical information, and allows the use of high spatial resolution and high contrast MR images to monitor and correct motion-corrupted PET data. In this paper we build on previous work to form a methodology for respiratory motion correction of PET data, and show it can improve PET image quality whilst having minimal impact on clinical PET-MR protocols. We introduce a joint PET-MR motion model, using only 1 min per PET bed position of simultaneously acquired PET and MR data to provide a respiratory motion correspondence model that captures inter-cycle and intra-cycle breathing variations. In the model setup, 2D multi-slice MR provides the dynamic imaging component, and PET data, via low spatial resolution framing and principal component analysis, provides the model surrogate. We evaluate different motion models (1D and 2D linear, and 1D and 2D polynomial) by computing model-fit and model-prediction errors on dynamic MR images on a data set of 45 patients. Finally we apply the motion model methodology to 5 clinical PET-MR oncology patient datasets. Qualitative PET reconstruction improvements and artefact reduction are assessed with visual analysis, and quantitative improvements are calculated using standardised uptake value (SUV(peak) and SUV(max)) changes in avid lesions. We demonstrate the capability of a joint PET-MR motion model to predict respiratory motion by showing significantly improved image quality of PET data acquired before the motion model data. The method can be used to incorporate motion into the reconstruction of any length of PET acquisition, with only 1 min of extra scan time, and with no external hardware required.

Non-small-cell lung cancer resectability: diagnostic value of PET/MR.

PURPOSE: To assess the diagnostic performance of PET/MR in patients with non-small-cell lung cancer. METHODS: Fifty consecutive consenting patients who underwent routine (18)F-FDG PET/CT for potentially radically treatable lung cancer following a staging CT scan were recruited for PET/MR imaging on the same day. Two experienced readers, unaware of the results with the other modalities, interpreted the PET/MR images independently. Discordances were resolved in consensus. PET/MR TNM staging was compared to surgical staging from thoracotomy as the reference standard in 33 patients. In the remaining 17 nonsurgical patients, TNM was determined based on histology from biopsy, imaging results (CT and PET/CT) and follow-up. ROC curve analysis was used to assess accuracy, sensitivity and specificity of the PET/MR in assessing the surgical resectability of primary tumour. The kappa statistic was used to assess interobserver agreement in the PET/MR TNM staging. Two different readers, without knowledge of the PET/MR findings, subsequently separately reviewed the PET/CT images for TNM staging. The generalized kappa statistic was used to determine intermodality agreement between PET/CT and PET/MR for TNM staging. RESULTS: ROC curve analysis showed that PET/MR had a specificity of 92.3 % and a sensitivity of 97.3 % in the determination of resectability with an AUC of 0.95. Interobserver agreement in PET/MR reading ranged from substantial to perfect between the two readers (Cohen's kappa 0.646 - 1) for T stage, N stage and M stage. Intermodality agreement between PET/CT and PET/MR ranged from substantial to almost perfect for T stage, N stage and M stage (Cohen's kappa 0.627 - 0.823). CONCLUSION: In lung cancer patients PET/MR appears to be a robust technique for preoperative staging.

Perfusion SPECT in cochlear implantation and promontory stimulation.

BACKGROUND: Recent studies of profoundly deaf patients with cochlear implants have demonstrated that these patients are able to process sound in the auditory cortex in a similar way to normal subjects. However, there are large variations in outcome. Various clinical criteria are used for subject selection and the decision as to which ear is to be implanted involves electrical stimulation of the promontory which is used to confirm the persistence of auditory neurones and fibres that can be utilized by the cochlear implant. In this study we have used SPECT with Tc-HMPAO to investigate activation of the auditory cortex in cochlear implantees post-surgery. In addition we also investigated whether electrical stimulation of the promontory does produce change in blood flow in the auditory cortex in pre-surgery candidates, which would indicate viable auditory networks that can be utilized by a cochlear implant device. METHODS AND RESULTS: Image analysis was performed with SPM99. Results of a simple subtraction paradigm indicated bilateral activation of auditory cortex and Wernicke's area in the post-implant group during auditory stimulus (speech) and bilateral activation of the ventral lateral posterior thalamus and bilateral auditory association cortex BA21/22/42, in the pre-implant group during electrical stimulus but no activation of the primary auditory cortex. A conjunction analysis used to investigate the common areas of activation across both groups during the stimulus condition showed that there was a common bilateral activation of the primary auditory cortex in both groups (BA22/41/42). In addition, analysis of a subset of the seven post-implant subjects who did not comprehend the speech in our study showed an activation (Pu<0.05, where Pu is the peak voxel threshold, uncorrected for multiple comparisons) in the left auditory cortex that extended into area BA22 synonymous with Wernicke's area. This supports the theory that this region has a sensory role.