Optimising brain age estimation through transfer learning: A suite of pre-trained foundation models for improved performance and generalisability in a clinical setting.

Estimated age from brain MRI data has emerged as a promising biomarker of neurological health. However, the absence of large, diverse, and clinically representative training datasets, along with the complexity of managing heterogeneous MRI data, presents significant barriers to the development of accurate and generalisable models appropriate for clinical use. Here, we present a deep learning framework trained on routine clinical data (N up to 18,890, age range 18-96 years). We trained five separate models for accurate brain age prediction (all with mean absolute error ≤4.0 years, R2 ≥ .86) across five different MRI sequences (T2 -weighted, T2 -FLAIR, T1 -weighted, diffusion-weighted, and gradient-recalled echo T2 *-weighted). Our trained models offer dual functionality. First, they have the potential to be directly employed on clinical data. Second, they can be used as foundation models for further refinement to accommodate a range of other MRI sequences (and therefore a range of clinical scenarios which employ such sequences). This adaptation process, enabled by transfer learning, proved effective in our study across a range of MRI sequences and scan orientations, including those which differed considerably from the original training datasets. Crucially, our findings suggest that this approach remains viable even with limited data availability (as low as N = 25 for fine-tuning), thus broadening the application of brain age estimation to more diverse clinical contexts and patient populations. By making these models publicly available, we aim to provide the scientific community with a versatile toolkit, promoting further research in brain age prediction and related areas.

Prevalence of Neuroradiological Abnormalities in First-Episode Psychosis: A Systematic Review and Meta-analysis.

Importance: Individuals presenting with first-episode psychosis (FEP) may have a secondary ("organic") etiology to their symptoms that can be identified using neuroimaging. Because failure to detect such cases at an early stage can have serious clinical consequences, it has been suggested that brain magnetic resonance imaging (MRI) should be mandatory for all patients presenting with FEP. However, this remains a controversial issue, partly because the prevalence of clinically relevant MRI abnormalities in this group is unclear. Objective: To derive a meta-analytic estimate of the prevalence of clinically relevant neuroradiological abnormalities in FEP. Data Sources: Electronic databases Ovid, MEDLINE, PubMed, Embase, PsychINFO, and Global Health were searched up to July 2021. References and citations of included articles and review articles were also searched. Study Selection: Magnetic resonance imaging studies of patients with FEP were included if they reported the frequency of intracranial radiological abnormalities. Data Extraction and Synthesis: Independent extraction was undertaken by 3 researchers and a random-effects meta-analysis of pooled proportions was calculated. Moderators were tested using subgroup and meta-regression analyses. Heterogeneity was evaluated using the I2 index. The robustness of results was evaluated using sensitivity analyses. Publication bias was assessed using funnel plots and Egger tests. Main Outcomes and Measures: Proportion of patients with a clinically relevant radiological abnormality (defined as a change in clinical management or diagnosis); number of patients needed to scan to detect 1 such abnormality (number needed to assess [NNA]). Results: Twelve independent studies (13 samples) comprising 1613 patients with FEP were included. Of these patients, 26.4% (95% CI, 16.3%-37.9%; NNA of 4) had an intracranial radiological abnormality, and 5.9% (95% CI, 3.2%-9.0%) had a clinically relevant abnormality, yielding an NNA of 18. There were high degrees of heterogeneity among the studies for these outcomes, 95% to 73%, respectively. The most common type of clinically relevant finding was white matter abnormalities, with a prevalence of 0.9% (95% CI, 0%-2.8%), followed by cysts, with a prevalence of 0.5% (95% CI, 0%-1.4%). Conclusions and Relevance: This systematic review and meta-analysis found that 5.9% of patients presenting with a first episode of psychosis had a clinically relevant finding on MRI. Because the consequences of not detecting these abnormalities can be serious, these findings support the use of MRI as part of the initial clinical assessment of all patients with FEP.

The clinical relevance of serum versus CSF NMDAR autoantibodies associated exclusively with psychiatric features: a systematic review and meta-analysis of individual patient data.

BACKGROUND: A variety of psychiatric syndromes are associated with NMDAR autoantibodies; however, their clinical relevance when only present in the serum is unclear. We explored whether patients with CSF NMDAR autoantibodies could be distinguished from patients with serum-only NMDAR autoantibodies. METHODS: The electronic databases MEDLINE, EMBASE, PubMed, and PsycINFO were searched. Articles reporting adult patients with isolated psychiatric features and positive for NMDAR autoantibodies with relevant investigations were included. Patient level meta-analysis compared patients positive for CSF NMDAR autoantibodies with patients positive for serum NMDAR autoantibodies, but negative for CSF NMDAR autoantibodies. Dichotomous data were analysed using crude odds ratios (OR), whilst continuous data were analysed using Mann-Whitney Test (U). The protocol was prospectively registered (CRD42018082210). RESULTS: Of 4413 publications, 42 were included, reporting 79 patients. Median age was 34 years (IQR 19 years); 56% (45/79) were female and 24% (16/68) had a tumour. In total, 41 patients were positive for CSF autoantibodies and 20 were positive for serum-only autoantibodies. Patients with CSF autoantibodies were significantly more likely to be female (p < 0.001) and have a rapid (< 3 month) onset of symptoms (p = 0.02) than patients with serum-only autoantibodies. They were also more likely to present with psychosis (p < 0.001), exhibit EEG (p = 0.006), MRI (p = 0.002), and CSF (p = 0.001) abnormalities, but less likely to present with insomnia (p = 0.04). CONCLUSIONS: Patients with an isolated psychiatric syndrome with CSF NMDAR autoantibodies can potentially be distinguished from those with serum-only NMDAR autoantibodies based on clinicodemographic and investigation findings.

Deep learning models for triaging hospital head MRI examinations.

The growing demand for head magnetic resonance imaging (MRI) examinations, along with a global shortage of radiologists, has led to an increase in the time taken to report head MRI scans in recent years. For many neurological conditions, this delay can result in poorer patient outcomes and inflated healthcare costs. Potentially, computer vision models could help reduce reporting times for abnormal examinations by flagging abnormalities at the time of imaging, allowing radiology departments to prioritise limited resources into reporting these scans first. To date, however, the difficulty of obtaining large, clinically-representative labelled datasets has been a bottleneck to model development. In this work, we present a deep learning framework, based on convolutional neural networks, for detecting clinically-relevant abnormalities in minimally processed, hospital-grade axial T2-weighted and axial diffusion-weighted head MRI scans. The models were trained at scale using a Transformer-based neuroradiology report classifier to generate a labelled dataset of 70,206 examinations from two large UK hospital networks, and demonstrate fast (< 5 s), accurate (area under the receiver operating characteristic curve (AUC) > 0.9), and interpretable classification, with good generalisability between hospitals (ΔAUC ≤ 0.02). Through a simulation study we show that our best model would reduce the mean reporting time for abnormal examinations from 28 days to 14 days and from 9 days to 5 days at the two hospital networks, demonstrating feasibility for use in a clinical triage environment.

Accurate brain-age models for routine clinical MRI examinations.

Convolutional neural networks (CNN) can accurately predict chronological age in healthy individuals from structural MRI brain scans. Potentially, these models could be applied during routine clinical examinations to detect deviations from healthy ageing, including early-stage neurodegeneration. This could have important implications for patient care, drug development, and optimising MRI data collection. However, existing brain-age models are typically optimised for scans which are not part of routine examinations (e.g., volumetric T1-weighted scans), generalise poorly (e.g., to data from different scanner vendors and hospitals etc.), or rely on computationally expensive pre-processing steps which limit real-time clinical utility. Here, we sought to develop a brain-age framework suitable for use during routine clinical head MRI examinations. Using a deep learning-based neuroradiology report classifier, we generated a dataset of 23,302 'radiologically normal for age' head MRI examinations from two large UK hospitals for model training and testing (age range = 18-95 years), and demonstrate fast (< 5 s), accurate (mean absolute error [MAE] < 4 years) age prediction from clinical-grade, minimally processed axial T2-weighted and axial diffusion-weighted scans, with generalisability between hospitals and scanner vendors (Δ MAE < 1 year). The clinical relevance of these brain-age predictions was tested using 228 patients whose MRIs were reported independently by neuroradiologists as showing atrophy 'excessive for age'. These patients had systematically higher brain-predicted age than chronological age (mean predicted age difference = +5.89 years, 'radiologically normal for age' mean predicted age difference = +0.05 years, p < 0.0001). Our brain-age framework demonstrates feasibility for use as a screening tool during routine hospital examinations to automatically detect older-appearing brains in real-time, with relevance for clinical decision-making and optimising patient pathways.

Deep learning to automate the labelling of head MRI datasets for computer vision applications.

OBJECTIVES: The purpose of this study was to build a deep learning model to derive labels from neuroradiology reports and assign these to the corresponding examinations, overcoming a bottleneck to computer vision model development. METHODS: Reference-standard labels were generated by a team of neuroradiologists for model training and evaluation. Three thousand examinations were labelled for the presence or absence of any abnormality by manually scrutinising the corresponding radiology reports ('reference-standard report labels'); a subset of these examinations (n = 250) were assigned 'reference-standard image labels' by interrogating the actual images. Separately, 2000 reports were labelled for the presence or absence of 7 specialised categories of abnormality (acute stroke, mass, atrophy, vascular abnormality, small vessel disease, white matter inflammation, encephalomalacia), with a subset of these examinations (n = 700) also assigned reference-standard image labels. A deep learning model was trained using labelled reports and validated in two ways: comparing predicted labels to (i) reference-standard report labels and (ii) reference-standard image labels. The area under the receiver operating characteristic curve (AUC-ROC) was used to quantify model performance. Accuracy, sensitivity, specificity, and F1 score were also calculated. RESULTS: Accurate classification (AUC-ROC > 0.95) was achieved for all categories when tested against reference-standard report labels. A drop in performance (ΔAUC-ROC > 0.02) was seen for three categories (atrophy, encephalomalacia, vascular) when tested against reference-standard image labels, highlighting discrepancies in the original reports. Once trained, the model assigned labels to 121,556 examinations in under 30 min. CONCLUSIONS: Our model accurately classifies head MRI examinations, enabling automated dataset labelling for downstream computer vision applications. KEY POINTS: • Deep learning is poised to revolutionise image recognition tasks in radiology; however, a barrier to clinical adoption is the difficulty of obtaining large labelled datasets for model training. • We demonstrate a deep learning model which can derive labels from neuroradiology reports and assign these to the corresponding examinations at scale, facilitating the development of downstream computer vision models. • We rigorously tested our model by comparing labels predicted on the basis of neuroradiology reports with two sets of reference-standard labels: (1) labels derived by manually scrutinising each radiology report and (2) labels derived by interrogating the actual images.

Management of chronic headache with referral from primary care to direct access to MRI compared with Neurology services: an observational prospective study in London.

OBJECTIVES: To evaluate the cost, accessibility and patient satisfaction implications of two clinical pathways used in the management of chronic headache. INTERVENTION: Management of chronic headache following referral from Primary Care that differed in the first appointment, either a Neurology appointment or an MRI brain scan. DESIGN AND SETTING: A pragmatic, non-randomised, prospective, single-centre study at a Central Hospital in London. PARTICIPANTS: Adult patients with chronic headache referred from primary to secondary care. PRIMARY AND SECONDARY OUTCOME MEASURES: Participants' use of healthcare services and costs were estimated using primary and secondary care databases and questionnaires quarterly up to 12 months postrecruitment. Cost analyses were compared using generalised linear models. Secondary outcomes assessed: access to care, patient satisfaction, headache burden and self-perceived quality of life using headache-specific (Migraine Disability Assessment Scale and Headache Impact Test) and a generic questionnaire (5-level EQ-5D). RESULTS: Mean (SD) cost up to 6 months postrecruitment per participant was £578 (£420) for the Neurology group (n=128) and £245 (£172) for the MRI group (n=95), leading to an estimated mean cost difference of £333 (95% CI £253 to £413, p<0.001). The mean cost difference at 12 months increased to £518 (95% CI £401 to £637, p<0.001). When adjusted for baseline and follow-up imbalances between groups, this remained statistically significant. The utilisation of brain MRI improved access to care compared with the Neurology group (p<0.001). Participants in the Neurology group reported higher levels of satisfaction associated with the pathway and led to greater change in care management. CONCLUSION: Direct referral to brain MRI from Primary Care led to cost-savings and quicker access to care but lower satisfaction levels when compared with referral to Neurology services. Further research into the use of brain MRI for a subset of patient population more likely to be reassured by a negative brain scan should be considered. TRIAL REGISTRATION NUMBER: NCT02753933.

Should GPs have direct access to imaging for headache? A qualitative study of patients' views in the UK.

OBJECTIVE: We aimed to describe patients' views of a new referral pathway of general practitioner (GP) direct access to MRI, versus imaging after referral to a specialist. DESIGN: This qualitative study involved 20 semistructured interviews. Twenty patients (10 from each pathway) were purposively recruited and interviewed to describe their attitudes. SETTING: A neurology headache clinic and neuroradiology services from the boroughs of Southwark and Lambeth in South London, UK. PARTICIPANTS: Twenty patients were involved in this study. RESULTS: Over half of the participants felt relieved once they received their scan results, while some remained uncertain about the underlying cause of their symptoms. Some participants described a long wait to see a specialist. Others described a long wait time to receive scan results, especially from their GP. Spontaneous reduction in headache symptoms occurred for some participants and for others, normal imaging results allowed them to focus more on symptom management. CONCLUSION: Relief was reported especially when scan results had been explained clearly and without too much delay. Those with continuing pain focused on how to get relief from symptoms. Patient experience might be improved with clearer information from GPs about how patients can access results, standard reporting procedures and closer liaison between neuroradiology and GPs.

Severe Respiratory Failure, Extracorporeal Membrane Oxygenation, and Intracranial Hemorrhage.

OBJECTIVES: For patients supported with veno-venous extracorporeal membrane oxygenation, the occurrence of intracranial hemorrhage is associated with a high mortality. It is unclear whether intracranial hemorrhage is a consequence of the extracorporeal intervention or of the underlying severe respiratory pathology. In a cohort of patients transferred to a regional severe respiratory failure center that routinely employs admission brain imaging, we sought 1) the prevalence of intracranial hemorrhage; 2) survival and neurologic outcomes; and 3) factors associated with intracranial hemorrhage. DESIGN: A single-center, retrospective, observational cohort study. SETTING: Tertiary referral severe respiratory failure center, university teaching hospital. PATIENTS: Patients admitted between December 2011 and February 2016. INTERVENTION: None. MEASUREMENTS AND MAIN RESULTS: Three hundred forty-two patients were identified: 250 managed with extracorporeal support and 92 managed using conventional ventilation. The prevalence of intracranial hemorrhage was 16.4% in extracorporeal membrane oxygenation patients and 7.6% in conventionally managed patients (p = 0.04). Multivariate analysis revealed factors independently associated with intracranial hemorrhage to be duration of ventilation (d) (odds ratio, 1.13 [95% CI, 1.03-1.23]; p = 0.011) and admission fibrinogen (g/L) (odds ratio, 0.73 [0.57-0.91]; p = 0.009); extracorporeal membrane oxygenation was not an independent risk factor (odds ratio, 3.29 [0.96-15.99]; p = 0.088). In patients who received veno-venous extracorporeal membrane oxygenation, there was no significant difference in 6-month survival between patients with and without intracranial hemorrhage (68.3% vs 76.0%; p = 0.350). Good neurologic function was observed in 92%. CONCLUSIONS: We report a higher prevalence of intracranial hemorrhage than has previously been described with high level of neurologically intact survival. Duration of mechanical ventilation and admission fibrinogen, but not exposure to extracorporeal support, are independently associated with intracranial hemorrhage.

Paediatric cerebrovascular CT angiography-towards better image quality.

BACKGROUND: Paediatric cerebrovascular CT angiography (CTA) can be challenging to perform due to variable cardiovascular physiology between different age groups and the risk of movement artefact. This analysis aimed to determine what proportion of CTA at our institution was of diagnostic quality and identify technical factors which could be improved. MATERIALS AND METHODS: a retrospective analysis of 20 cases was performed at a national paediatric neurovascular centre assessing image quality with a subjective scoring system and Hounsfield Unit (HU) measurements. Demographic data, contrast dose, flow rate and triggering times were recorded for each patient. RESULTS: Using a qualitative scoring system, 75% of studies were found to be of diagnostic quality (n=9 'good', n=6 'satisfactory') and 25% (n=5) were 'poor'. Those judged subjectively to be poor had arterial contrast density measured at less than 250 HU. Increased arterial opacification was achieved for cases performed with an increased flow rate (2.5-4 mL/s) and higher intravenous contrast dose (2 mL/kg). Triggering was found to be well timed in nine cases, early in four cases and late in seven cases. Of the scans triggered early, 75% were poor. Of the scans triggered late, less (29%) were poor. CONCLUSIONS: High flow rates (>2.5 mL/s) were a key factor for achieving high quality paediatric cerebrovascular CTA imaging. However, appropriate triggering by starting the scan immediately on contrast opacification of the monitoring vessel plays an important role and could maintain image quality when flow rates were lower. Early triggering appeared more detrimental than late.

Imaging of degenerative lumbar intervertebral discs; linking anatomy, pathology and imaging.

Low back pain is a common medical condition that has significant implications for healthcare providers and the UK economy. Low back pain can be classified as 'specific' in which an underlying pathophysiological mechanism is identified (eg, herniated intervertebral disc). Advanced imaging should be performed in this situation and in those patients in whom systemic disease is strongly suspected. In the majority (approximately 90%), low back pain in 'non specific' and there is a weak correlation with imaging abnormalities. This is an area of ongoing research and remains controversial in terms of imaging approach and treatment (eg, theory of discogenic pain, interpretation and treatment of endplate changes). With regards Modic endplate changes, current research suggests that an infective component may be involved that may identify novel potential treatments in patients with chronic low back pain refractory to other treatment modalities. MRI is the imaging modality of choice for the assessment of degenerative changes in intervertebral discs. MRI has superior soft tissue contrast resolution when compared to other imaging modalities (eg, plain radiography, CT). An understanding of normal anatomy and MR appearances of intervertebral discs, particularly with regards to how these appearances change with advancing age, is required to aid image interpretation. Knowledge of the spectrum of degenerative processes that may occur in the intervertebral discs is required in order to identify and explain abnormal MRI appearances. As the communication of MRI findings may guide therapeutic decision making and surgical intervention, the terminology used by radiologists must be accurate and consistent. Therefore, description of degenerative disc changes in the current paper is based on the most up-to-date recommendations, the aim being to aid reporting by radiologists and interpretation of reports by referring clinicians.