The virtual reference radiologist: comprehensive AI assistance for clinical image reading and interpretation.

OBJECTIVES: Large language models (LLMs) have shown potential in radiology, but their ability to aid radiologists in interpreting imaging studies remains unexplored. We investigated the effects of a state-of-the-art LLM (GPT-4) on the radiologists' diagnostic workflow. MATERIALS AND METHODS: In this retrospective study, six radiologists of different experience levels read 40 selected radiographic [n = 10], CT [n = 10], MRI [n = 10], and angiographic [n = 10] studies unassisted (session one) and assisted by GPT-4 (session two). Each imaging study was presented with demographic data, the chief complaint, and associated symptoms, and diagnoses were registered using an online survey tool. The impact of Artificial Intelligence (AI) on diagnostic accuracy, confidence, user experience, input prompts, and generated responses was assessed. False information was registered. Linear mixed-effect models were used to quantify the factors (fixed: experience, modality, AI assistance; random: radiologist) influencing diagnostic accuracy and confidence. RESULTS: When assessing if the correct diagnosis was among the top-3 differential diagnoses, diagnostic accuracy improved slightly from 181/240 (75.4%, unassisted) to 188/240 (78.3%, AI-assisted). Similar improvements were found when only the top differential diagnosis was considered. AI assistance was used in 77.5% of the readings. Three hundred nine prompts were generated, primarily involving differential diagnoses (59.1%) and imaging features of specific conditions (27.5%). Diagnostic confidence was significantly higher when readings were AI-assisted (p > 0.001). Twenty-three responses (7.4%) were classified as hallucinations, while two (0.6%) were misinterpretations. CONCLUSION: Integrating GPT-4 in the diagnostic process improved diagnostic accuracy slightly and diagnostic confidence significantly. Potentially harmful hallucinations and misinterpretations call for caution and highlight the need for further safeguarding measures. CLINICAL RELEVANCE STATEMENT: Using GPT-4 as a virtual assistant when reading images made six radiologists of different experience levels feel more confident and provide more accurate diagnoses; yet, GPT-4 gave factually incorrect and potentially harmful information in 7.4% of its responses.

(+)-[18F]Flubatine as a novel α4ß2 nicotinic acetylcholine receptor PET ligand-results of the first-in-human brain imaging application in patients with ß-amyloid PET-confirmed Alzheimer's disease and healthy controls.

PURPOSES: We present the first in-human brain PET imaging data of the new α4ß2 nicotinic acetylcholine receptor (nAChR)-targeting radioligand (+)-[18F]Flubatine. Aims were to develop a kinetic modeling-based approach to quantify (+)-[18F]Flubatine and compare the data of healthy controls (HCs) and patients with Alzheimer's disease (AD); to investigate the partial volume effect (PVE) on regional (+)-[18F]Flubatine binding; and whether (+)-[18F]Flubatine binding and cognitive test data respective ß-amyloid radiotracer accumulation were correlated. METHODS: We examined 11 HCs and 9 mild AD patients. All subjects underwent neuropsychological testing and [11C]PiB PET/MRI examination. (+)-[18F]Flubatine PET data were evaluated using full kinetic modeling and regional as well as voxel-based analyses. RESULTS: With 270-min p.i., the unchanged parent compound amounted to 97 ± 2%. Adequate fits of the time-activity curves were obtained with the 1 tissue compartment model (1TCM). (+)-[18F]Flubatine distribution volume (binding) was significantly reduced in bilateral mesial temporal cortex in AD patients compared with HCs (right 10.6 ± 1.1 vs 11.6 ± 1.4, p = 0.049; left 11.0 ± 1.1 vs 12.2 ± 1.8, p = 0.046; one-sided t tests each). PVE correction increased not only (+)-[18F]Flubatine binding of approximately 15% but also standard deviation of 0.4-70%. Cognitive test data and (+)-[18F]Flubatine binding were significantly correlated in the left anterior cingulate, right posterior cingulate, and right parietal cortex (r > 0.5, p < 0.05 each). In AD patients, (+)-[18F]Flubatine binding and [11C]PiB standardized uptake value ratios were negatively correlated in several regions; whereas in HCs, a positive correlation between cortical (+)-[18F]Flubatine binding and [11C]PiB accumulation in the white matter was found. No adverse event related to (+)-[18F]Flubatine occurred. CONCLUSION: (+)-[18F]Flubatine is a safe and stable PET ligand. Full kinetic modeling can be realized by 1TCM without metabolite correction. (+)-[18F]Flubatine binding affinity was high enough to detect group differences. Of interest, correlation between white matter ß-amyloid PET uptake and (+)-[18F]Flubatine binding indicated an association between white matter integrity and availability of α4ß2 nAChRs. Overall, (+)-[18F]Flubatine showed favorable characteristics and has therefore the potential to serve as α4ß2 nAChR-targeting PET ligand in further clinical trials.

Cognitive correlates of α4ß2 nicotinic acetylcholine receptors in mild Alzheimer's dementia.

In early Alzheimer's dementia, there is a need for PET biomarkers of disease progression with close associations to cognitive dysfunction that may aid to predict further cognitive decline and neurodegeneration. Amyloid biomarkers are not suitable for that purpose. The α4ß2 nicotinic acetylcholine receptors (α4ß2-nAChRs) are widely abundant in the human brain. As neuromodulators they play an important role in cognitive functions such as attention, learning and memory. Post-mortem studies reported lower expression of α4ß2-nAChRs in more advanced Alzheimer's dementia. However, there is ongoing controversy whether α4ß2-nAChRs are reduced in early Alzheimer's dementia. Therefore, using the recently developed α4ß2-nAChR-specific radioligand (-)-18F-flubatine and PET, we aimed to quantify the α4ß2-nAChR availability and its relationship to specific cognitive dysfunction in mild Alzheimer's dementia. Fourteen non-smoking patients with mild Alzheimer's dementia, drug-naïve for cholinesterase therapy, were compared with 15 non-smoking healthy controls matched for age, sex and education by applying (-)-18F-flubatine PET together with a neuropsychological test battery. The one-tissue compartment model and Logan plot method with arterial input function were used for kinetic analysis to obtain the total distribution volume (VT) as the primary, and the specific binding part of the distribution volume (VS) as the secondary quantitative outcome measure of α4ß2-nAChR availability. VS was determined by using a pseudo-reference region. Correlations between VT within relevant brain regions and Z-scores of five cognitive functions (episodic memory, executive function/working memory, attention, language, visuospatial function) were calculated. VT (and VS) were applied for between-group comparisons. Volume of interest and statistical parametric mapping analyses were carried out. Analyses revealed that in patients with mild Alzheimer's dementia compared to healthy controls, there was significantly lower VT, especially within the hippocampus, fronto-temporal cortices, and basal forebrain, which was similar to comparisons of VS. VT decline in Alzheimer's dementia was associated with distinct domains of impaired cognitive functioning, especially episodic memory and executive function/working memory. Using (-)-18F-flubatine PET in patients with mild Alzheimer's dementia, we show for the first time a cholinergic α4ß2-nAChR deficiency mainly present within the basal forebrain-cortical and septohippocampal cholinergic projections and a relationship between lower α4ß2-nAChR availability and impairment of distinct cognitive domains, notably episodic memory and executive function/working memory. This shows the potential of (-)-18F-flubatine as PET biomarker of cholinergic α4ß2-nAChR dysfunction and specific cognitive decline. Thus, if validated by longitudinal PET studies, (-)-18F-flubatine might become a PET biomarker of progression of neurodegeneration in Alzheimer's dementia.

Radiation dosimetry of the α4ß2 nicotinic receptor ligand (+)-[18F]flubatine, comparing preclinical PET/MRI and PET/CT to first-in-human PET/CT results.

BACKGROUND: Both enantiomers of [18F]flubatine are new radioligands for neuroimaging of α4ß2 nicotinic acetylcholine receptors with positron emission tomography (PET) exhibiting promising pharmacokinetics which makes them attractive for different clinical questions. In a previous preclinical study, the main advantage of (+)-[18F]flubatine compared to (-)-[18F]flubatine was its higher binding affinity suggesting that (+)-[18F]flubatine might be able to detect also slight reductions of α4ß2 nAChRs and could be more sensitive than (-)-[18F]flubatine in early stages of Alzheimer's disease. To support the clinical translation, we investigated a fully image-based internal dosimetry approach for (+)-[18F]flubatine, comparing mouse data collected on a preclinical PET/MRI system to piglet and first-in-human data acquired on a clinical PET/CT system. Time-activity curves (TACs) were obtained from the three species, the animal data extrapolated to human scale, exponentially fitted and the organ doses (OD), and effective dose (ED) calculated with OLINDA. RESULTS: The excreting organs (urinary bladder, kidneys, and liver) receive the highest organ doses in all species. Hence, a renal/hepatobiliary excretion pathway can be assumed. In addition, the ED conversion factors of 12.1 µSv/MBq (mice), 14.3 µSv/MBq (piglets), and 23.0 µSv/MBq (humans) were calculated which are well within the order of magnitude as known from other 18F-labeled radiotracers. CONCLUSIONS: Although both enantiomers of [18F]flubatine exhibit different binding kinetics in the brain due to the respective affinities, the effective dose revealed no enantiomer-specific differences among the investigated species. The preclinical dosimetry and biodistribution of (+)-[18F]flubatine was shown and the feasibility of a dose assessment based on image data acquired on a small animal PET/MR and a clinical PET/CT was demonstrated. Additionally, the first-in-human study confirmed the tolerability of the radiation risk of (+)-[18F]flubatine imaging which is well within the range as caused by other 18F-labeled tracers. However, as shown in previous studies, the ED in humans is underestimated by up to 50 % using preclinical imaging for internal dosimetry. This fact needs to be considered when applying for first-in-human studies based on preclinical biokinetic data scaled to human anatomy.

First-in-human PET quantification study of cerebral α4ß2* nicotinic acetylcholine receptors using the novel specific radioligand (-)-[(18)F]Flubatine.

α4ß2* nicotinic receptors (α4ß2* nAChRs) could provide a biomarker in neuropsychiatric disorders (e.g., Alzheimer's and Parkinson's diseases, depressive disorders, and nicotine addiction). However, there is a lack of α4ß2* nAChR specific PET radioligands with kinetics fast enough to enable quantification of nAChR within a reasonable time frame. Following on from promising preclinical results, the aim of the present study was to evaluate for the first time in humans the novel PET radioligand (-)-[(18)F]Flubatine, formerly known as (-)-[(18)F]NCFHEB, as a tool for α4ß2* nAChR imaging and in vivo quantification. Dynamic PET emission recordings lasting 270min were acquired on an ECAT EXACT HR+ scanner in 12 healthy male non-smoking subjects (71.0±5.0years) following the intravenous injection of 353.7±9.4MBq of (-)-[(18)F]Flubatine. Individual magnetic resonance imaging (MRI) was performed for co-registration. PET frames were motion-corrected, before the kinetics in 29 brain regions were characterized using 1- and 2-tissue compartment models (1TCM, 2TCM). Given the low amounts of metabolite present in plasma, we tested arterial input functions with and without metabolite corrections. In addition, pixel-based graphical analysis (Logan plot) was used. The model's goodness of fit, with and without metabolite correction was assessed by Akaike's information criterion. Model parameters of interest were the total distribution volume VT (mL/cm(3)), and the binding potential BPND relative to the corpus callosum, which served as a reference region. The tracer proved to have high stability in vivo, with 90% of the plasma radioactivity remaining as untransformed parent compound at 90min, fast brain kinetics with rapid uptake and equilibration between free and receptor-bound tracer. Adequate fits of brain TACs were obtained with the 1TCM. VT could be reliably estimated within 90min for all regions investigated, and within 30min for low-binding regions such as the cerebral cortex. The rank order of VT by region corresponded well with the known distribution of α4ß2* receptors (VT [thalamus] 27.4±3.8, VT [putamen] 12.7±0.9, VT [frontal cortex] 10.0±0.8, and VT [corpus callosum] 6.3±0.8). The BPND, which is a parameter of α4ß2* nAChR availability, was 3.41±0.79 for the thalamus, 1.04±0.25 for the putamen and 0.61±0.23 for the frontal cortex, indicating high specific tracer binding. Use of the arterial input function without metabolite correction resulted in a 10% underestimation in VT, and was without important biasing effects on BPND. Altogether, kinetics and imaging properties of (-)-[(18)F]Flubatine appear favorable and suggest that (-)-[(18)F]Flubatine is a very suitable and clinically applicable PET tracer for in vivo imaging of α4ß2* nAChRs in neuropsychiatric disorders.

Internal dose assessment of (-)-18F-flubatine, comparing animal model datasets of mice and piglets with first-in-human results.

UNLABELLED: (-)-(18)F-flubatine is a promising tracer for neuroimaging of nicotinic acetylcholine receptors (nAChRs), subtype α4ß2, using PET. Radiation doses after intravenous administration of the tracer in mice and piglets were assessed to determine the organ doses (ODs) and the effective dose (ED) to humans. The results were compared with subsequent clinical investigations in human volunteers. METHODS: Twenty-seven female CD1 mice (weight ± SD, 28.2 ± 2.1 g) received intravenous injection of 0.75 ± 0.33 MBq of (-)-(18)F-flubatine. Up to 240 min after injection, 3 animals per time point were sacrificed and the organs harvested, weighed, and counted in a γ counter to determine mass and activity, respectively. Furthermore, whole-body PET scans of 5 female piglets (age ± SD, 44 ± 3 d; weight ± SD, 13.7 ± 1.7 kg) and 3 humans (2 men and 1 woman; age ± SD, 59.6 ± 3.9 y; weight ± SD, 74.3 ± 3.1 kg) were obtained up to 236 min (piglets) and 355 min (humans) after injection of 186.6 ± 7.4 and 353.7 ± 10.2 MBq of (-)-(18)F-flubatine, respectively, using a PET/CT scanner. The CT was used for delineation of the organs. Exponential curves were fitted to the time-activity-data, and time and mass scales were adapted to the human anatomy. The ODs were calculated using OLINDA/EXM (version 1.0); EDs were calculated with the tissue-weighting factors of ICRP103. RESULTS: After the injection of (-)-(18)F-flubatine, there were no adverse or clinically detectable pharmacologic effects in any of the subjects. The highest activities after injection were found in the kidneys, urinary bladder, and liver. The urinary bladder receives the highest OD in all investigated species, followed by the kidneys and the liver for animals and humans, respectively. On the basis of mouse, piglet, and human kinetic data, the projected human ED of (-)-(18)F-flubatine was estimated to be 12.5 µSv/MBq in mice, 14.7 ± 0.7 µSv/MBq in piglets, and 23.4 ± 0.4 µSv/MBq in humans. CONCLUSION: As has been demonstrated for other PET radiotracers, preclinical (i.e., animal-derived) dosimetry underestimates the ED to humans, in the current case of (-)-(18)F-flubatine by 34%-44%.

Evaluation of metabolism, plasma protein binding and other biological parameters after administration of (-)-[(18)F]Flubatine in humans.

INTRODUCTION: (-)-[(18)F]Flubatine is a PET tracer with high affinity and selectivity for the nicotinic acetylcholine α4ß2 receptor subtype. A clinical trial assessing the availability of this subtype of nAChRs was performed. From a total participant number of 21 Alzheimer's disease (AD) patients and 20 healthy controls (HCs), the following parameters were determined: plasma protein binding, metabolism and activity distribution between plasma and whole blood. METHODS: Plasma protein binding and fraction of unchanged parent compound were assessed by ultracentrifugation and HPLC, respectively. The distribution of radioactivity (parent compound+metabolites) between plasma and whole blood was determined ex vivo at different time-points after injection by gamma counting after separation of whole blood by centrifugation into the cellular and non-cellular components. In additional experiments in vitro, tracer distribution between these blood components was assessed for up to 90min. RESULTS: A fraction of 15%±2% of (-)-[(18)F]Flubatine was found to be bound to plasma proteins. Metabolic degradation of (-)-[(18)F]Flubatine was very low, resulting in almost 90% unchanged parent compound at 90min p.i. with no significant difference between AD and HC. The radioactivity distribution between plasma and whole blood changed in vivo only slightly over time from 0.82±0.03 at 3min p.i. to 0.87±0.03 at 270min p.i. indicating the contribution of only a small amount of metabolites. In vitro studies revealed that (-)-[(18)F]Flubatine was instantaneously distributed between cellular and non-cellular blood parts. DISCUSSION: (-)-[(18)F]Flubatine exhibits very favourable characteristics for a PET radiotracer such as slow metabolic degradation and moderate plasma protein binding. Equilibrium of radioactivity distribution between plasma and whole blood is reached instantaneously and remains almost constant over time allowing both convenient sample handling and facilitated fractional blood volume contribution assessment.

Fully automated radiosynthesis of both enantiomers of [18F]Flubatine under GMP conditions for human application.

A fully automatized radiosynthesis of (+)- and (-)-[(18)F]Flubatine ((+)- and (-)NCFHEB) by means of a commercially available synthesis module (TRACERlab FX FN) under GMP conditions is reported. Radiochemical yields of 30% within an overall synthesis time of 40 min were achieved in more than 70 individual syntheses. Specific activities were approximately 3000 GBq/µmol and radiochemical purity was determined to be at least 97%.