Large Language Models for Automated Synoptic Reports and Resectability Categorization in Pancreatic Cancer.

Background Structured radiology reports for pancreatic ductal adenocarcinoma (PDAC) improve surgical decision-making over free-text reports, but radiologist adoption is variable. Resectability criteria are applied inconsistently. Purpose To evaluate the performance of large language models (LLMs) in automatically creating PDAC synoptic reports from original reports and to explore performance in categorizing tumor resectability. Materials and Methods In this institutional review board-approved retrospective study, 180 consecutive PDAC staging CT reports on patients referred to the authors' European Society for Medical Oncology-designated cancer center from January to December 2018 were included. Reports were reviewed by two radiologists to establish the reference standard for 14 key findings and National Comprehensive Cancer Network (NCCN) resectability category. GPT-3.5 and GPT-4 (accessed September 18-29, 2023) were prompted to create synoptic reports from original reports with the same 14 features, and their performance was evaluated (recall, precision, F1 score). To categorize resectability, three prompting strategies (default knowledge, in-context knowledge, chain-of-thought) were used for both LLMs. Hepatopancreaticobiliary surgeons reviewed original and artificial intelligence (AI)-generated reports to determine resectability, with accuracy and review time compared. The McNemar test, t test, Wilcoxon signed-rank test, and mixed effects logistic regression models were used where appropriate. Results GPT-4 outperformed GPT-3.5 in the creation of synoptic reports (F1 score: 0.997 vs 0.967, respectively). Compared with GPT-3.5, GPT-4 achieved equal or higher F1 scores for all 14 extracted features. GPT-4 had higher precision than GPT-3.5 for extracting superior mesenteric artery involvement (100% vs 88.8%, respectively). For categorizing resectability, GPT-4 outperformed GPT-3.5 for each prompting strategy. For GPT-4, chain-of-thought prompting was most accurate, outperforming in-context knowledge prompting (92% vs 83%, respectively; P = .002), which outperformed the default knowledge strategy (83% vs 67%, P < .001). Surgeons were more accurate in categorizing resectability using AI-generated reports than original reports (83% vs 76%, respectively; P = .03), while spending less time on each report (58%; 95% CI: 0.53, 0.62). Conclusion GPT-4 created near-perfect PDAC synoptic reports from original reports. GPT-4 with chain-of-thought achieved high accuracy in categorizing resectability. Surgeons were more accurate and efficient using AI-generated reports. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Chang in this issue.

A large normative connectome for exploring the tractographic correlates of focal brain interventions.

Diffusion-weighted MRI (dMRI) is a widely used neuroimaging modality that permits the in vivo exploration of white matter connections in the human brain. Normative structural connectomics - the application of large-scale, group-derived dMRI datasets to out-of-sample cohorts - have increasingly been leveraged to study the network correlates of focal brain interventions, insults, and other regions-of-interest (ROIs). Here, we provide a normative, whole-brain connectome in MNI space that enables researchers to interrogate fiber streamlines that are likely perturbed by given ROIs, even in the absence of subject-specific dMRI data. Assembled from multi-shell dMRI data of 985 healthy Human Connectome Project subjects using generalized Q-sampling imaging and multispectral normalization techniques, this connectome comprises ~12 million unique streamlines, the largest to date. It has already been utilized in at least 18 peer-reviewed publications, most frequently in the context of neuromodulatory interventions like deep brain stimulation and focused ultrasound. Now publicly available, this connectome will constitute a useful tool for understanding the wider impact of focal brain perturbations on white matter architecture going forward.

Use of GPT-4 With Single-Shot Learning to Identify Incidental Findings in Radiology Reports.

GPT-4 identified incidental adrenal nodules, pancreatic cystic lesions, and vascular calcifications in radiology reports with F1 scores of 1.00, 0.91, and 0.99, respectively. The findings indicate a potential role for large language models to help improve recognition and management of incidental imaging findings and to be applied flexibly in a medical context.

Successful magnetic resonance-guided focused ultrasound treatment of tremor in patients with a skull density ratio of 0.4 or less.

OBJECTIVE: The use of magnetic resonance-guided focused ultrasound (MRgFUS) for the treatment of tremor-related disorders and other novel indications has been limited by guidelines advocating treatment of patients with a skull density ratio (SDR) above 0.45 ± 0.05 despite reports of successful outcomes in patients with a low SDR (LSDR). The authors' goal was to retrospectively analyze the sonication strategies, adverse effects, and clinical and imaging outcomes in patients with SDR ≤ 0.4 treated for tremor using MRgFUS. METHODS: Clinical outcomes and adverse effects were assessed at 3 and 12 months after MRgFUS. Outcomes and lesion location, volume, and shape characteristics (elongation and eccentricity) were compared between the SDR groups. RESULTS: A total of 102 consecutive patients were included in the analysis, of whom 39 had SDRs ≤ 0.4. No patient was excluded from treatment because of an LSDR, with the lowest being 0.22. Lesioning temperatures (> 52°C) and therapeutic ablations were achieved in all patients. There were no significant differences in clinical outcome, adverse effects, lesion location, and volume between the high SDR group and the LSDR group. SDR was significantly associated with total energy (rho = -0.459, p < 0.001), heating efficiency (rho = 0.605, p < 0.001), and peak temperature (rho = 0.222, p = 0.025). CONCLUSIONS: The authors' results show that treatment of tremor in patients with an LSDR using MRgFUS is technically possible, leading to a safe and lasting therapeutic effect. Limiting the number of sonications and adjusting the energy and duration to achieve the required temperature early during the treatment are suitable strategies in LSDR patients.

Uncovering neuroanatomical correlates of impaired coordinated movement after pallidal deep brain stimulation.

BACKGROUND: The loss of the ability to swim following deep brain stimulation (DBS), although rare, poses a worrisome risk of drowning. It is unclear what anatomic substrate and neural circuitry underlie this phenomenon. We report a case of cervical dystonia with lost ability to swim and dance during active stimulation of globus pallidus internus. We investigated the anatomical underpinning of this phenomenon using unique functional and structural imaging analysis. METHODS: Tesla (3T) functional MRI (fMRI) of the patient was used during active DBS and compared with a cohort of four matched patients without this side effect. Structural connectivity mapping was used to identify brain network engagement by stimulation. RESULTS: fMRI during stimulation revealed significant (Pbonferroni<0.0001) stimulation-evoked responses (DBS ON

Comparative neural correlates of DBS and MRgFUS lesioning for tremor control in essential tremor.

BACKGROUND: Given high rates of early complications and non-reversibility, refined targeting is necessitated for magnetic resonance-guided focused ultrasound (MRgFUS) thalamotomy for essential tremor (ET). Selection of lesion location can be informed by considering optimal stimulation area from deep brain stimulation (DBS). METHODS: 118 patients with ET who received DBS (39) or MRgFUS (79) of the ventral intermediate nucleus (VIM) underwent stimulation/lesion mapping, probabilistic mapping of clinical efficacy and normative structural connectivity analysis. The efficacy maps were compared, which depict the relationship between stimulation/lesion location and clinical outcome. RESULTS: Efficacy maps overlap around the VIM ventral border and encompass the dentato-rubro-thalamic tract. While the MRgFUS map extends inferiorly into the posterior subthalamic area, the DBS map spreads inside the VIM antero-superiorly. CONCLUSION: Comparing the efficacy maps of DBS and MRgFUS suggests a potential alternative location for lesioning, more antero-superiorly. This may reduce complications, without sacrificing efficacy, and individualise targeting. TRIAL REGISTRATION NUMBER: NCT02252380.

WarpDrive: Improving spatial normalization using manual refinements.

Spatial normalization-the process of mapping subject brain images to an average template brain-has evolved over the last 20+ years into a reliable method that facilitates the comparison of brain imaging results across patients, centers & modalities. While overall successful, sometimes, this automatic process yields suboptimal results, especially when dealing with brains with extensive neurodegeneration and atrophy patterns, or when high accuracy in specific regions is needed. Here we introduce WarpDrive, a novel tool for manual refinements of image alignment after automated registration. We show that the tool applied in a cohort of patients with Alzheimer's disease who underwent deep brain stimulation surgery helps create more accurate representations of the data as well as meaningful models to explain patient outcomes. The tool is built to handle any type of 3D imaging data, also allowing refinements in high-resolution imaging, including histology and multiple modalities to precisely aggregate multiple data sources together.

Dissociable default-mode subnetworks subserve childhood attention and cognitive flexibility: Evidence from deep learning and stereotactic electroencephalography.

Cognitive flexibility encompasses the ability to efficiently shift focus and forms a critical component of goal-directed attention. The neural substrates of this process are incompletely understood in part due to difficulties in sampling the involved circuitry. We leverage stereotactic intracranial recordings to directly resolve local-field potentials from otherwise inaccessible structures to study moment-to-moment attentional activity in children with epilepsy performing a flexible attentional task. On an individual subject level, we employed deep learning to decode neural features predictive of task performance indexed by single-trial reaction time. These models were subsequently aggregated across participants to identify predictive brain regions based on AAL atlas and FIND functional network parcellations. Through this approach, we show that fluctuations in beta (12-30 Hz) and gamma (30-80 Hz) power reflective of increased top-down attentional control and local neuronal processing within relevant large-scale networks can accurately predict single-trial task performance. We next performed connectomic profiling of these highly predictive nodes to examine task-related engagement of distributed functional networks, revealing exclusive recruitment of the dorsal default mode network during shifts in attention. The identification of distinct substreams within the default mode system supports a key role for this network in cognitive flexibility and attention in children. Furthermore, convergence of our results onto consistent functional networks despite significant inter-subject variability in electrode implantations supports a broader role for deep learning applied to intracranial electrodes in the study of human attention.

Local neuroanatomical and tract-based proxies of optimal subcallosal cingulate deep brain stimulation.

BACKGROUND: Deep brain stimulation of the subcallosal cingulate area (SCC-DBS) is a promising neuromodulatory therapy for treatment-resistant depression (TRD). Biomarkers of optimal target engagement are needed to guide surgical targeting and stimulation parameter selection and to reduce variance in clinical outcome. OBJECTIVE/HYPOTHESIS: We aimed to characterize the relationship between stimulation location, white matter tract engagement, and clinical outcome in a large (n = 60) TRD cohort treated with SCC-DBS. A smaller cohort (n = 22) of SCC-DBS patients with differing primary indications (bipolar disorder/anorexia nervosa) was utilized as an out-of-sample validation cohort. METHODS: Volumes of tissue activated (VTAs) were constructed in standard space using high-resolution structural MRI and individual stimulation parameters. VTA-based probabilistic stimulation maps (PSMs) were generated to elucidate voxelwise spatial patterns of efficacious stimulation. A whole-brain tractogram derived from Human Connectome Project diffusion-weighted MRI data was seeded with VTA pairs, and white matter streamlines whose overlap with VTAs related to outcome ('discriminative' streamlines; Puncorrected < 0.05) were identified using t-tests. Linear modelling was used to interrogate the potential clinical relevance of VTA overlap with specific structures. RESULTS: PSMs varied by hemisphere: high-value left-sided voxels were located more anterosuperiorly and squarely in the lateral white matter, while the equivalent right-sided voxels fell more posteroinferiorly and involved a greater proportion of grey matter. Positive discriminative streamlines localized to the bilateral (but primarily left) cingulum bundle, forceps minor/rostrum of corpus callosum, and bilateral uncinate fasciculus. Conversely, negative discriminative streamlines mostly belonged to the right cingulum bundle and bilateral uncinate fasciculus. The best performing linear model, which utilized information about VTA volume overlap with each of the positive discriminative streamline bundles as well as the negative discriminative elements of the right cingulum bundle, explained significant variance in clinical improvement in the primary TRD cohort (R = 0.46, P < 0.001) and survived repeated 10-fold cross-validation (R = 0.50, P = 0.040). This model was also able to predict outcome in the out-of-sample validation cohort (R = 0.43, P = 0.047). CONCLUSION(S): These findings reinforce prior indications of the importance of white matter engagement to SCC-DBS treatment success while providing new insights that could inform surgical targeting and stimulation parameter selection decisions.

Multi-centre analysis of networks and genes modulated by hypothalamic stimulation in patients with aggressive behaviours.

Deep brain stimulation targeting the posterior hypothalamus (pHyp-DBS) is being investigated as a treatment for refractory aggressive behavior, but its mechanisms of action remain elusive. We conducted an integrated imaging analysis of a large multi-centre dataset, incorporating volume of activated tissue modeling, probabilistic mapping, normative connectomics, and atlas-derived transcriptomics. Ninety-one percent of the patients responded positively to treatment, with a more striking improvement recorded in the pediatric population. Probabilistic mapping revealed an optimized surgical target within the posterior-inferior-lateral region of the posterior hypothalamic area. Normative connectomic analyses identified fiber tracts and functionally connected with brain areas associated with sensorimotor function, emotional regulation, and monoamine production. Functional connectivity between the target, periaqueductal gray and key limbic areas - together with patient age - were highly predictive of treatment outcome. Transcriptomic analysis showed that genes involved in mechanisms of aggressive behavior, neuronal communication, plasticity and neuroinflammation might underlie this functional network.

Optimal deep brain stimulation sites and networks for stimulation of the fornix in Alzheimer's disease.

Deep brain stimulation (DBS) to the fornix is an investigational treatment for patients with mild Alzheimer's Disease. Outcomes from randomized clinical trials have shown that cognitive function improved in some patients but deteriorated in others. This could be explained by variance in electrode placement leading to differential engagement of neural circuits. To investigate this, we performed a post-hoc analysis on a multi-center cohort of 46 patients with DBS to the fornix (NCT00658125, NCT01608061). Using normative structural and functional connectivity data, we found that stimulation of the circuit of Papez and stria terminalis robustly associated with cognitive improvement (R = 0.53, p < 0.001). On a local level, the optimal stimulation site resided at the direct interface between these structures (R = 0.48, p < 0.001). Finally, modulating specific distributed brain networks related to memory accounted for optimal outcomes (R = 0.48, p < 0.001). Findings were robust to multiple cross-validation designs and may define an optimal network target that could refine DBS surgery and programming.

Magnetic resonance-guided focused ultrasound for the treatment of tremor.

INTRODUCTION: Magnetic resonance-guided focused ultrasound (MRgFUS) is an emerging treatment for tremor and other movement disorders. An incisionless therapy, it is becoming increasingly common worldwide. However, given MRgFUS' relative novelty, there remain limited data on its benefits and adverse effects. AREAS COVERED: We review the current state of evidence of MRgFUS for tremor, highlight its challenges, and discuss future perspectives. EXPERT OPINION: Essential tremor (ET) has been the major indication for MRgFUS since a milestone randomized controlled trial (RCT) in 2016, with substantial evidence attesting to the efficacy and acceptable safety profile of this treatment. Patients with other tremor etiologies are also being treated with MRgFUS, with studies - including an RCT - suggesting parkinsonian tremor in particular responds well to this intervention. Additionally, targets other than the ventral intermediate nucleus, such as the subthalamic nucleus and internal segment of the globus pallidus, have been reported to improve parkinsonian symptoms beyond tremor, including rigidity and bradykinesia. Although MRgFUS is encumbered by certain unique technical challenges, it nevertheless offers significant advantages compared to alternative neurosurgical interventions for tremor. The fast-growing interest in this treatment modality will likely lead to further scientific and technological advancements that could optimize and expand its therapeutic potential.

Ipsilateral and axial tremor response to focused ultrasound thalamotomy for essential tremor: clinical outcomes and probabilistic mapping.

BACKGROUND: MR-guided focused ultrasound (MRgFUS) thalamotomy has been shown to be a safe and effective treatment for essential tremor (ET). OBJECTIVE: To investigate the effects of MRgFUS in patients with ET with an emphasis on ipsilateral-hand and axial tremor subscores. METHODS: Tremor scores and adverse effects of 100 patients treated between 2012 and 2018 were assessed at 1 week, 3, 12, and 24 months. A subgroup analysis of ipsilateral-hand tremor responders (defined as patients with ≥30% improvement at any time point) and non-responders was performed. Correlations and predictive factors for improvement were analysed. Weighted probabilistic maps of improvement were generated. RESULTS: Significant improvement in axial, contralateral-hand and total tremor scores was observed at all study visits from baseline (p<0.0001). There was no significant improvement in ipsilateral subscores. A subset of patients (n=20) exhibited group-level ipsilateral-hand improvement that remained significant through all follow-ups (p<0.001). Multivariate regression analysis revealed that higher baseline scores predict better improvement in ipsilateral-hand and axial tremor. Probabilistic maps demonstrated that the lesion hotspot for axial improvement was situated more medially than that for contralateral improvement. CONCLUSION: MRgFUS significantly improved axial, contralateral-hand and total tremor scores. In a subset of patients, a consistent group-level treatment effect was observed for ipsilateral-hand tremor. While ipsilateral improvement seemed to be less directly related to lesion location, a spatial relationship between lesion location and axial and contralateral improvement was observed that proved consistent with the somatotopic organisation of the ventral intermediate nucleus. TRIAL REGISTRATION NUMBERS: NCT01932463, NCT01827904, and NCT02252380.

Neural Correlates of Optimal Deep Brain Stimulation for Cervical Dystonia.

A total of 15 individuals with cervical dystonia and good outcome after pallidal deep brain stimulation underwent resting-state functional magnetic resonance imaging under three conditions: stimulation using a priori clinically determined optimal settings (ON-Op), non-optimal settings (ON-NOp), and stimulation off (OFF). ON-Op > OFF and ON-Op > ON-NOp were both associated with significant deactivation within sensorimotor cortex (changes not seen with ON-NOp > OFF). Brain responses to stimulation were related to individual long-term clinical improvement (R = 0.73, R2 = 0.53, p = 0.001). The relationship was consistent when this model included four additional patients with generalized or truncal dystonia. These findings highlight the potential for immediate imaging-based biomarkers of clinical efficacy. ANN NEUROL 2022;92:418-424.

Identifying the neural network for neuromodulation in epilepsy through connectomics and graphs.

Deep brain stimulation is a treatment option for patients with drug-resistant epilepsy. The precise mechanism of neuromodulation in epilepsy is unknown, and biomarkers are needed for optimizing treatment. The aim of this study was to describe the neural network associated with deep brain stimulation targets for epilepsy and to explore its potential application as a novel biomarker for neuromodulation. Using seed-to-voxel functional connectivity maps, weighted by seizure outcomes, brain areas associated with stimulation were identified in normative resting state functional scans of 1000 individuals. To pinpoint specific regions in the normative epilepsy deep brain stimulation network, we examined overlapping areas of functional connectivity between the anterior thalamic nucleus, centromedian thalamic nucleus, hippocampus and less studied epilepsy deep brain stimulation targets. Graph network analysis was used to describe the relationship between regions in the identified network. Furthermore, we examined the associations of the epilepsy deep brain stimulation network with disease pathophysiology, canonical resting state networks and findings from a systematic review of resting state functional MRI studies in epilepsy deep brain stimulation patients. Cortical nodes identified in the normative epilepsy deep brain stimulation network were in the anterior and posterior cingulate, medial frontal and sensorimotor cortices, frontal operculum and bilateral insulae. Subcortical nodes of the network were in the basal ganglia, mesencephalon, basal forebrain and cerebellum. Anterior thalamic nucleus was identified as a central hub in the network with the highest betweenness and closeness values, while centromedian thalamic nucleus and hippocampus showed average centrality values. The caudate nucleus and mammillothalamic tract also displayed high centrality values. The anterior cingulate cortex was identified as an important cortical hub associated with the effect of deep brain stimulation in epilepsy. The neural network of deep brain stimulation targets shared hubs with known epileptic networks and brain regions involved in seizure propagation and generalization. Two cortical clusters identified in the epilepsy deep brain stimulation network included regions corresponding to resting state networks, mainly the default mode and salience networks. Our results were concordant with findings from a systematic review of resting state functional MRI studies in patients with deep brain stimulation for epilepsy. Our findings suggest that the various epilepsy deep brain stimulation targets share a common cortico-subcortical network, which might in part underpin the antiseizure effects of stimulation. Interindividual differences in this network functional connectivity could potentially be used as biomarkers in selection of patients, stimulation parameters and neuromodulation targets.

Unilateral Campotomy of Forel for Acquired Hemidystonia: An Open-Label Clinical Trial.

BACKGROUND: Hemidystonia (HD) is characterized by unilateral involuntary torsion movements and fixed postures of the limbs and face. It often develops after deleterious neuroplastic changes secondary to injuries to the brain. This condition usually responds poorly to medical treatment, and deep brain stimulation often yields unsatisfactory results. We propose this study based on encouraging results from case reports of patients with HD treated by ablative procedures in the subthalamic region. OBJECTIVE: To compare the efficacy of stereotactic-guided radiofrequency lesioning of the subthalamic area vs available medical treatment in patients suffering from acquired HD. METHODS: This is an open-label study in patients with secondary HD allocated according to their treatment choice, either surgical or medical treatment; both groups were followed for one year. Patients assigned in the surgical group underwent unilateral campotomy of Forel. The efficacy was assessed using the Unified Dystonia Rating Scale, Fahn-Marsden Dystonia Scale, Arm Dystonia Disability Scale, and SF-36 questionnaire scores. RESULTS: Patients in the surgical group experienced significant improvement in the Unified Dystonia Rating Scale, Fahn-Marsden Dystonia Scale, and Arm Dystonia Disability Scale (39%, 35%, and 15%, respectively) 1 year after the surgery, with positive reflex in quality-of-life measures, such as bodily pain and role-emotional process. Patients kept on medical treatment did not experience significant changes during the follow-up. No infections were recorded, and no neurological adverse events were associated with either intervention. CONCLUSION: The unilateral stereotaxy-guided ablation of Forel H1 and H2 fields significantly improved in patients with HD compared with optimized clinical treatment.