A retrieval-augmented chatbot based on GPT-4 provides appropriate differential diagnosis in gastrointestinal radiology: a proof of concept study.

BACKGROUND: We investigated the potential of an imaging-aware GPT-4-based chatbot in providing diagnoses based on imaging descriptions of abdominal pathologies. METHODS: Utilizing zero-shot learning via the LlamaIndex framework, GPT-4 was enhanced using the 96 documents from the Radiographics Top 10 Reading List on gastrointestinal imaging, creating a gastrointestinal imaging-aware chatbot (GIA-CB). To assess its diagnostic capability, 50 cases on a variety of abdominal pathologies were created, comprising radiological findings in fluoroscopy, MRI, and CT. We compared the GIA-CB to the generic GPT-4 chatbot (g-CB) in providing the primary and 2 additional differential diagnoses, using interpretations from senior-level radiologists as ground truth. The trustworthiness of the GIA-CB was evaluated by investigating the source documents as provided by the knowledge-retrieval mechanism. Mann-Whitney U test was employed. RESULTS: The GIA-CB demonstrated a high capability to identify the most appropriate differential diagnosis in 39/50 cases (78%), significantly surpassing the g-CB in 27/50 cases (54%) (p = 0.006). Notably, the GIA-CB offered the primary differential in the top 3 differential diagnoses in 45/50 cases (90%) versus g-CB with 37/50 cases (74%) (p = 0.022) and always with appropriate explanations. The median response time was 29.8 s for GIA-CB and 15.7 s for g-CB, and the mean cost per case was $0.15 and $0.02, respectively. CONCLUSIONS: The GIA-CB not only provided an accurate diagnosis for gastrointestinal pathologies, but also direct access to source documents, providing insight into the decision-making process, a step towards trustworthy and explainable AI. Integrating context-specific data into AI models can support evidence-based clinical decision-making. RELEVANCE STATEMENT: A context-aware GPT-4 chatbot demonstrates high accuracy in providing differential diagnoses based on imaging descriptions, surpassing the generic GPT-4. It provided formulated rationale and source excerpts supporting the diagnoses, thus enhancing trustworthy decision-support. KEY POINTS: • Knowledge retrieval enhances differential diagnoses in a gastrointestinal imaging-aware chatbot (GIA-CB). • GIA-CB outperformed the generic counterpart, providing formulated rationale and source excerpts. • GIA-CB has the potential to pave the way for AI-assisted decision support systems.

Cerebral microstructural alterations in Post-COVID-condition are related to cognitive impairment, olfactory dysfunction and fatigue.

After contracting COVID-19, a substantial number of individuals develop a Post-COVID-Condition, marked by neurologic symptoms such as cognitive deficits, olfactory dysfunction, and fatigue. Despite this, biomarkers and pathophysiological understandings of this condition remain limited. Employing magnetic resonance imaging, we conduct a comparative analysis of cerebral microstructure among patients with Post-COVID-Condition, healthy controls, and individuals that contracted COVID-19 without long-term symptoms. We reveal widespread alterations in cerebral microstructure, attributed to a shift in volume from neuronal compartments to free fluid, associated with the severity of the initial infection. Correlating these alterations with cognition, olfaction, and fatigue unveils distinct affected networks, which are in close anatomical-functional relationship with the respective symptoms.

Non-invasive biomarkers for spontaneous intracranial hypotension (SIH) through phase-contrast MRI.

BACKGROUND AND OBJECTIVE: Spontaneous intracranial hypotension (SIH) is an underdiagnosed disease. To depict the accurate diagnosis can be demanding; especially the detection of CSF-venous fistulas poses many challenges. Potential dynamic biomarkers have been identified through non-invasive phase-contrast MRI in a limited subset of SIH patients with evidence of spinal longitudinal extradural collection. This study aimed to explore these biomarkers related to spinal cord motion and CSF velocities in a broader SIH cohort. METHODS: A retrospective, monocentric pooled-data analysis was conducted of patients suspected to suffer from SIH who underwent phase-contrast MRI for spinal cord and CSF velocity measurements at segment C2/C3 referred to a tertiary center between February 2022 and June 2023. Velocity ranges (mm/s), total displacement (mm), and further derivatives were assessed and compared to data from the database of 70 healthy controls. RESULTS: In 117 patients, a leak was located (54% ventral leak, 20% lateral leak, 20% CSF-venous fistulas, 6% sacral leaks). SIH patients showed larger spinal cord and CSF velocities than healthy controls: e.g., velocity range 7.6 ± 3 mm/s vs. 5.6 ± 1.4 mm/s, 56 ± 21 mm/s vs. 42 ± 10 mm/s, p < 0.001, respectively. Patients with lateral leaks and CSF-venous fistulas exhibited an exceptionally heightened level of spinal cord motion (e.g., velocity range 8.4 ± 3.3 mm/s; 8.2 ± 3.1 mm/s vs. 5.6 ± 1.4 mm/s, p < 0.001, respectively). CONCLUSION: Phase-contrast MRI might become a valuable tool for SIH diagnosis, especially in patients with CSF-venous fistulas without evidence of spinal extradural fluid collection.

Quantification of cervical spinal stenosis by automated 3D MRI segmentation of spinal cord and cerebrospinal fluid space.

DESIGN: Prospective diagnostic study. OBJECTIVES: Anatomical evaluation and graduation of the severity of spinal stenosis is essential in degenerative cervical spine disease. In clinical practice, this is subjectively categorized on cervical MRI lacking an objective and reliable classification. We implemented a fully-automated quantification of spinal canal compromise through 3D T2-weighted MRI segmentation. SETTING: Medical Center - University of Freiburg, Germany. METHODS: Evaluation of 202 participants receiving 3D T2-weighted MRI of the cervical spine. Segments C2/3 to C6/7 were analyzed for spinal cord and cerebrospinal fluid space volume through a fully-automated segmentation based on a trained deep convolutional neural network. Spinal canal narrowing was characterized by relative values, across sever segments as adapted Maximal Canal Compromise (aMCC), and within the index segment as adapted Spinal Cord Occupation Ratio (aSCOR). Additionally, all segments were subjectively categorized by three observers as "no", "relative" or "absolute" stenosis. Computed scores were applied on the subjective categorization. RESULTS: 798 (79.0%) segments were subjectively categorized as "no" stenosis, 85 (8.4%) as "relative" stenosis, and 127 (12.6%) as "absolute" stenosis. The calculated scores revealed significant differences between each category (p ≤ 0.001). Youden's Index analysis of ROC curves revealed optimal cut-offs to distinguish between "no" and "relative" stenosis for aMCC = 1.18 and aSCOR = 36.9%, and between "relative" and "absolute" stenosis for aMCC = 1.54 and aSCOR = 49.3%. CONCLUSION: The presented fully-automated segmentation algorithm provides high diagnostic accuracy and objective classification of cervical spinal stenosis. The calculated cut-offs can be used for convenient radiological quantification of the severity of spinal canal compromise in clinical routine.

Photon-counting detector computed tomography for metal artifact reduction: a comparative study of different artifact reduction techniques in patients with orthopedic implants.

PURPOSE: Artifacts caused by metallic implants remain a challenge in computed tomography (CT). We investigated the impact of photon-counting detector computed tomography (PCD-CT) for artifact reduction in patients with orthopedic implants with respect to image quality and diagnostic confidence using different artifact reduction approaches. MATERIAL AND METHODS: In this prospective study, consecutive patients with orthopedic implants underwent PCD-CT imaging of the implant area. Four series were reconstructed for each patient (clinical standard reconstruction [PCD-CTStd], monoenergetic images at 140 keV [PCD-CT140keV], iterative metal artifact reduction (iMAR) corrected [PCD-CTiMAR], combination of iMAR and 140 keV monoenergetic [PCD-CT140keV+iMAR]). Subsequently, three radiologists evaluated the reconstructions in a random and blinded manner for image quality, artifact severity, anatomy delineation (adjacent and distant), and diagnostic confidence using a 5-point Likert scale (5 = excellent). In addition, the coefficient of variation [CV] and the relative quantitative artifact reduction potential were obtained as objective measures. RESULTS: We enrolled 39 patients with a mean age of 67.3 ± 13.2 years (51%; n = 20 male) and a mean BMI of 26.1 ± 4 kg/m2. All image quality measures and diagnostic confidence were significantly higher for the iMAR vs. non-iMAR reconstructions (all p < 0.001). No significant effect of the different artifact reduction approaches on CV was observed (p = 0.26). The quantitative analysis indicated the most effective artifact reduction for the iMAR reconstructions, which was higher than PCD-CT140keV (p < 0.001). CONCLUSION: PCD-CT allows for effective metal artifact reduction in patients with orthopedic implants, resulting in superior image quality and diagnostic confidence with the potential to improve patient management and clinical decision making.

The connection of motor improvement after deep brain stimulation in Parkinson's disease and microstructural integrity of the substantia nigra and subthalamic nucleus.

BACKGROUND: Nigrostriatal microstructural integrity has been suggested as a biomarker for levodopa response in Parkinson's disease (PD), which is a strong predictor for motor response to deep brain stimulation (DBS) of the subthalamic nucleus (STN). This study aimed to explore the impact of microstructural integrity of the substantia nigra (SN), STN, and putamen on motor response to STN-DBS using diffusion microstructure imaging. METHODS: Data was collected from 23 PD patients (mean age 63 ± 7, 6 females) who underwent STN-DBS, had preoperative 3 T diffusion magnetic resonance imaging including multishell diffusion-weighted MRI with b-values of 1000 and 2000 s/mm2 and records of motor improvement available. RESULTS: The association between a poorer DBS-response and increased free interstitial fluid showed notable effect sizes (rho > |0.4|) in SN and STN, but not in putamen. However, this did not reach significance after Bonferroni correction and controlling for sex and age. CONCLUSION: Microstructural integrity of SN and STN are potential biomarkers for the prediction of therapy efficacy following STN-DBS, but further studies are required to confirm these associations.

Association Between Body Composition and Survival in Patients With Gastroesophageal Adenocarcinoma: An Automated Deep Learning Approach.

PURPOSE: Body composition (BC) may play a role in outcome prognostication in patients with gastroesophageal adenocarcinoma (GEAC). Artificial intelligence provides new possibilities to opportunistically quantify BC from computed tomography (CT) scans. We developed a deep learning (DL) model for fully automatic BC quantification on routine staging CTs and determined its prognostic role in a clinical cohort of patients with GEAC. MATERIALS AND METHODS: We developed and tested a DL model to quantify BC measures defined as subcutaneous and visceral adipose tissue (VAT) and skeletal muscle on routine CT and investigated their prognostic value in a cohort of patients with GEAC using baseline, 3-6-month, and 6-12-month postoperative CTs. Primary outcome was all-cause mortality, and secondary outcome was disease-free survival (DFS). Cox regression assessed the association between (1) BC at baseline and mortality and (2) the decrease in BC between baseline and follow-up scans and mortality/DFS. RESULTS: Model performance was high with Dice coefficients ≥0.94 ± 0.06. Among 299 patients with GEAC (age 63.0 ± 10.7 years; 19.4% female), 140 deaths (47%) occurred over a median follow-up of 31.3 months. At baseline, no BC measure was associated with DFS. Only a substantial decrease in VAT >70% after a 6- to 12-month follow-up was associated with mortality (hazard ratio [HR], 1.99 [95% CI, 1.18 to 3.34]; P = .009) and DFS (HR, 1.73 [95% CI, 1.01 to 2.95]; P = .045) independent of age, sex, BMI, Union for International Cancer Control stage, histologic grading, resection status, neoadjuvant therapy, and time between surgery and follow-up CT. CONCLUSION: DL enables opportunistic estimation of BC from routine staging CT to quantify prognostic information. In patients with GEAC, only a substantial decrease of VAT 6-12 months postsurgery was an independent predictor for DFS beyond traditional risk factors, which may help to identify individuals at high risk who go otherwise unnoticed.

Imaging Markers Derived From MRI-Based Automated Kidney Segmentation­an Analysis of Data From the German National Cohort (NAKO Gesundheitsstudie)

BACKGROUND: Population-wide research on potential new imaging biomarkers of the kidney depends on accurate automated segmentation of the kidney and its compartments (cortex, medulla, and sinus). METHODS: We developed a robust deep-learning framework for kidney (sub-)segmentation based on a hierarchical, three-dimensional convolutional neural network (CNN) that was optimized for multi-scale problems of combined localization and segmentation. We applied the CNN to abdominal magnetic resonance images from the population-based German National Cohort (NAKO) study. RESULTS: There was good to excellent agreement between the model predictions and manual segmentations. The median values for the body-surface normalized total kidney, cortex, medulla, and sinus volumes of 9934 persons were 158, 115, 43, and 24 mL/m2. Distributions of these markers are provided both for the overall study population and for a subgroup of persons without kidney disease or any associated conditions. Multivariable adjusted regression analyses revealed that diabetes, male sex, and a higher estimated glomerular filtration rate (eGFR) are important predictors of higher total and cortical volumes. Each increase of eGFR by one unit (i.e., 1 mL/min per 1.73 m2 body surface area) was associated with a 0.98 mL/m2 increase in total kidney volume, and this association was significant. Volumes were lower in persons with eGFR-defined chronic kidney disease. CONCLUSION: The extraction of image-based biomarkers through CNN-based renal sub-segmentation using data from a population-based study yields reliable results, forming a solid foundation for future investigations.

Impact of temporal resolution on perfusion metrics, therapy decision, and radiation dose reduction in brain CT perfusion in patients with suspected stroke.

PURPOSE: CT perfusion of the brain is a powerful tool in stroke imaging, though the radiation dose is rather high. Several strategies for dose reduction have been proposed, including increasing the intervals between the dynamic scans. We determined the impact of temporal resolution on perfusion metrics, therapy decision, and radiation dose reduction in brain CT perfusion from a large dataset of patients with suspected stroke. METHODS: We retrospectively included 3555 perfusion scans from our clinical routine dataset. All cases were processed using the perfusion software VEOcore with a standard sampling of 1.5 s, as well as simulated reduced temporal resolution of 3.0, 4.5, and 6.0 s by leaving out respective time points. The resulting perfusion maps and calculated volumes of infarct core and mismatch were compared quantitatively. Finally, hypothetical decisions for mechanical thrombectomy following the DEFUSE-3 criteria were compared. RESULTS: The agreement between calculated volumes for core (ICC = 0.99, 0.99, and 0.98) and hypoperfusion (ICC = 0.99, 0.99, and 0.97) was excellent for all temporal sampling schemes. Of the 1226 cases with vascular occlusion, 14 (1%) for 3.0 s sampling, 23 (2%) for 4.5 s sampling, and 63 (5%) for 6.0 s sampling would have been treated differently if the DEFUSE-3 criteria had been applied. Reduction of temporal resolution to 3.0 s, 4.5 s, and 6.0 s reduced the radiation dose by a factor of 2, 3, or 4. CONCLUSION: Reducing the temporal sampling of brain perfusion CT has only a minor impact on image quality and treatment decision, but significantly reduces the radiation dose to that of standard non-contrast CT.

Accuracy of automated segmentation and volumetry of acute intracerebral hemorrhage following minimally invasive surgery using a patch-based convolutional neural network in a small dataset.

PURPOSE: In cases of acute intracerebral hemorrhage (ICH) volume estimation is of prognostic and therapeutic value following minimally invasive surgery (MIS). The ABC/2 method is widely used, but suffers from inaccuracies and is time consuming. Supervised machine learning using convolutional neural networks (CNN), trained on large datasets, is suitable for segmentation tasks in medical imaging. Our objective was to develop a CNN based machine learning model for the segmentation of ICH and of the drain and volumetry of ICH following MIS of acute supratentorial ICH on a relatively small dataset. METHODS: Ninety two scans were assigned to training (n = 29 scans), validation (n = 4 scans) and testing (n = 59 scans) datasets. The mean age (SD) was 70 (± 13.56) years. Male patients were 36. A hierarchical, patch-based CNN for segmentation of ICH and drain was trained. Volume of ICH was calculated from the segmentation mask. RESULTS: The best performing model achieved a Dice similarity coefficient of 0.86 and 0.91 for the ICH and drain respectively. Automated ICH volumetry yielded high agreement with ground truth (Intraclass correlation coefficient = 0.94 [95% CI: 0.91, 0.97]). Average difference in the ICH volume was 1.33 mL. CONCLUSION: Using a relatively small dataset, originating from different CT-scanners and with heterogeneous voxel dimensions, we applied a patch-based CNN framework and successfully developed a machine learning model, which accurately segments the intracerebral hemorrhage (ICH) and the drains. This provides automated and accurate volumetry of the bleeding in acute ICH treated with minimally invasive surgery.

Improving the use of LLMs in radiology through prompt engineering: from precision prompts to zero-shot learning.

PURPOSE: Large language models (LLMs) such as ChatGPT have shown significant potential in radiology. Their effectiveness often depends on prompt engineering, which optimizes the interaction with the chatbot for accurate results. Here, we highlight the critical role of prompt engineering in tailoring the LLMs' responses to specific medical tasks. MATERIALS AND METHODS: Using a clinical case, we elucidate different prompting strategies to adapt the LLM ChatGPT using GPT4 to new tasks without additional training of the base model. These approaches range from precision prompts to advanced in-context methods such as few-shot and zero-shot learning. Additionally, the significance of embeddings, which serve as a data representation technique, is discussed. RESULTS: Prompt engineering substantially improved and focused the chatbot's output. Moreover, embedding of specialized knowledge allows for more transparent insight into the model's decision-making and thus enhances trust. CONCLUSION: Despite certain challenges, prompt engineering plays a pivotal role in harnessing the potential of LLMs for specialized tasks in the medical domain, particularly radiology. As LLMs continue to evolve, techniques like few-shot learning, zero-shot learning, and embedding-based retrieval mechanisms will become indispensable in delivering tailored outputs. KEY POINTS: · Large language models might impact radiological practice and decision-masking.. · However, implementation and performance are dependent on the assigned task.. · Optimization of prompting strategies can substantially improve model performance.. · Strategies for prompt engineering range from precision prompts to zero-shot learning..

A deep learning approach for projection and body-side classification in musculoskeletal radiographs.

BACKGROUND: The growing prevalence of musculoskeletal diseases increases radiologic workload, highlighting the need for optimized workflow management and automated metadata classification systems. We developed a large-scale, well-characterized dataset of musculoskeletal radiographs and trained deep learning neural networks to classify radiographic projection and body side. METHODS: In this IRB-approved retrospective single-center study, a dataset of musculoskeletal radiographs from 2011 to 2019 was retrieved and manually labeled for one of 45 possible radiographic projections and the depicted body side. Two classification networks were trained for the respective tasks using the Xception architecture with a custom network top and pretrained weights. Performance was evaluated on a hold-out test sample, and gradient-weighted class activation mapping (Grad-CAM) heatmaps were computed to visualize the influential image regions for network predictions. RESULTS: A total of 13,098 studies comprising 23,663 radiographs were included with a patient-level dataset split, resulting in 19,183 training, 2,145 validation, and 2,335 test images. Focusing on paired body regions, training for side detection included 16,319 radiographs (13,284 training, 1,443 validation, and 1,592 test images). The models achieved an overall accuracy of 0.975 for projection and 0.976 for body-side classification on the respective hold-out test sample. Errors were primarily observed in projections with seamless anatomical transitions or non-orthograde adjustment techniques. CONCLUSIONS: The deep learning neural networks demonstrated excellent performance in classifying radiographic projection and body side across a wide range of musculoskeletal radiographs. These networks have the potential to serve as presorting algorithms, optimizing radiologic workflow and enhancing patient care. RELEVANCE STATEMENT: The developed networks excel at classifying musculoskeletal radiographs, providing valuable tools for research data extraction, standardized image sorting, and minimizing misclassifications in artificial intelligence systems, ultimately enhancing radiology workflow efficiency and patient care. KEY POINTS: • A large-scale, well-characterized dataset was developed, covering a broad spectrum of musculoskeletal radiographs. • Deep learning neural networks achieved high accuracy in classifying radiographic projection and body side. • Grad-CAM heatmaps provided insight into network decisions, contributing to their interpretability and trustworthiness. • The trained models can help optimize radiologic workflow and manage large amounts of data.

Reducing False-Positives in CT Perfusion Infarct Core Segmentation Using Contralateral Local Normalization.

BACKGROUND AND PURPOSE: The established global threshold of rCBF <30% for infarct core segmentation can lead to false-positives, as it does not account for the differences in blood flow between GM and WM and patient-individual factors, such as microangiopathy. To mitigate this problem, we suggest normalizing each voxel not only with a global reference value (ie, the median value of normally perfused tissue) but also with its local contralateral counterpart. MATERIALS AND METHODS: We retrospectively enrolled 2830 CTP scans with suspected ischemic stroke, of which 335 showed obvious signs of microangiopathy. In addition to the conventional, global normalization, a local normalization was performed by dividing the rCBF maps with their mirrored and smoothed counterpart, which sets each voxel value in relation to the contralateral counterpart, intrinsically accounting for GM and WM differences and symmetric patient individual microangiopathy. Maps were visually assessed and core volumes were calculated for both methods. RESULTS: Cases with obvious microangiopathy showed a strong reduction in false-positives by using local normalization (mean 14.7 mL versus mean 3.7 mL in cases with and without microangiopathy). On average, core volumes were slightly smaller, indicating an improved segmentation that was more robust against naturally low blood flow values in the deep WM. CONCLUSIONS: The proposed method of local normalization can reduce overestimation of the infarct core, especially in the deep WM and in cases with obvious microangiopathy. False-positives in CTP infarct core segmentation might lead to less-than-optimal therapy decisions when not correctly interpreted. The proposed method might help mitigate this problem.

Disbalanced recruitment of crossed and uncrossed cerebello-thalamic pathways during deep brain stimulation is predictive of delayed therapy escape in essential tremor.

BACKGROUND: Thalamic deep brain stimulation (DBS) is an efficacious treatment for drug-resistant essential tremor (ET) and the dentato-rubro-thalamic tract (DRT) constitutes an important target structure. However, up to 40% of patients habituate and lose treatment efficacy over time, frequently accompanied by a stimulation-induced cerebellar syndrome. The phenomenon termed delayed therapy escape (DTE) is insufficiently understood. Our previous work showed that DTE clinically is pronounced on the non-dominant side and suggested that differential involvement of crossed versus uncrossed DRT (DRTx/DRTu) might play a role in DTE development. METHODS: We retrospectively enrolled right-handed patients under bilateral thalamic DBS >12 months for ET from a cross-sectional study. They were characterized with the Fahn-Tolosa-Marin Tremor Rating Scale (FTMTRS) and Scale for the Assessment and Rating of Ataxia (SARA) scores at different timepoints. Normative fiber tractographic evaluations of crossed and uncrossed cerebellothalamic pathways and volume of activated tissue (VAT) studies together with [18F]Fluorodeoxyglucose positron emission tomography were applied. RESULTS: A total of 29 patients met the inclusion criteria. Favoring DRTu over DRTx in the non-dominant VAT was associated with DTE (R2 = 0.4463, p < 0.01) and ataxia (R2 = 0.2319, p < 0.01). Moreover, increasing VAT size on the right (non-dominant) side was associated at trend level with more asymmetric glucose metabolism shifting towards the right (dominant) dentate nucleus. CONCLUSION: Our results suggest that a disbalanced recruitment of DRTu in the non-dominant VAT induces detrimental stimulation effects on the dominant cerebellar outflow (together with contralateral stimulation) leading to DTE and thus hampering the overall treatment efficacy.

A content-aware chatbot based on GPT 4 provides trustworthy recommendations for Cone-Beam CT guidelines in dental imaging.

OBJECTIVES: To develop a content-aware chatbot based on GPT-3.5-Turbo and GPT-4 with specialized knowledge on the German S2 Cone-Beam CT (CBCT) dental imaging guideline and to compare the performance against humans. METHODS: The LlamaIndex software library was used to integrate the guideline context into the chatbots. Based on the CBCT S2 guideline, 40 questions were posed to content-aware chatbots and early career and senior practitioners with different levels of experience served as reference. The chatbots' performance was compared in terms of recommendation accuracy and explanation quality. Chi-square test and one-tailed Wilcoxon signed rank test evaluated accuracy and explanation quality, respectively. RESULTS: The GPT-4 based chatbot provided 100% correct recommendations and superior explanation quality compared to the one based on GPT3.5-Turbo (87.5% vs. 57.5% for GPT-3.5-Turbo; P = .003). Moreover, it outperformed early career practitioners in correct answers (P = .002 and P = .032) and earned higher trust than the chatbot using GPT-3.5-Turbo (P = 0.006). CONCLUSIONS: A content-aware chatbot using GPT-4 reliably provided recommendations according to current consensus guidelines. The responses were deemed trustworthy and transparent, and therefore facilitate the integration of artificial intelligence into clinical decision-making.

Deep learning to estimate impaired glucose metabolism from Magnetic Resonance Imaging of the liver: An opportunistic population screening approach.

AIM: Diabetes is a global health challenge, and many individuals are undiagnosed and not aware of their increased risk of morbidity/mortality although dedicated tests are available, which indicates the need for novel population-wide screening approaches. Here, we developed a deep learning pipeline for opportunistic screening of impaired glucose metabolism using routine magnetic resonance imaging (MRI) of the liver and tested its prognostic value in a general population setting. METHODS: In this retrospective study a fully automatic deep learning pipeline was developed to quantify liver shape features on routine MR imaging using data from a prospective population study. Subsequently, the association between liver shape features and impaired glucose metabolism was investigated in individuals with prediabetes, type 2 diabetes and healthy controls without prior cardiovascular diseases. K-medoids clustering (3 clusters) with a dissimilarity matrix based on Euclidean distance and ordinal regression was used to assess the association between liver shape features and glycaemic status. RESULTS: The deep learning pipeline showed a high performance for liver shape analysis with a mean Dice score of 97.0±0.01. Out of 339 included individuals (mean age 56.3±9.1 years; males 58.1%), 79 (23.3%) and 46 (13.6%) were classified as having prediabetes and type 2 diabetes, respectively. Individuals in the high risk cluster using all liver shape features (n = 14) had a 2.4 fold increased risk of impaired glucose metabolism after adjustment for cardiometabolic risk factors (age, sex, BMI, total cholesterol, alcohol consumption, hypertension, smoking and hepatic steatosis; OR 2.44 [95% CI 1.12-5.38]; p = 0.03). Based on individual shape features, the strongest association was found between liver volume and impaired glucose metabolism after adjustment for the same risk factors (OR 1.97 [1.38-2.85]; p<0.001). CONCLUSIONS: Deep learning can estimate impaired glucose metabolism on routine liver MRI independent of cardiometabolic risk factors and hepatic steatosis.

AI-based X-ray fracture analysis of the distal radius: accuracy between representative classification, detection and segmentation deep learning models for clinical practice.

OBJECTIVES: To aid in selecting the optimal artificial intelligence (AI) solution for clinical application, we directly compared performances of selected representative custom-trained or commercial classification, detection and segmentation models for fracture detection on musculoskeletal radiographs of the distal radius by aligning their outputs. DESIGN AND SETTING: This single-centre retrospective study was conducted on a random subset of emergency department radiographs from 2008 to 2018 of the distal radius in Germany. MATERIALS AND METHODS: An image set was created to be compatible with training and testing classification and segmentation models by annotating examinations for fractures and overlaying fracture masks, if applicable. Representative classification and segmentation models were trained on 80% of the data. After output binarisation, their derived fracture detection performances as well as that of a standard commercially available solution were compared on the remaining X-rays (20%) using mainly accuracy and area under the receiver operating characteristic (AUROC). RESULTS: A total of 2856 examinations with 712 (24.9%) fractures were included in the analysis. Accuracies reached up to 0.97 for the classification model, 0.94 for the segmentation model and 0.95 for BoneView. Cohen's kappa was at least 0.80 in pairwise comparisons, while Fleiss' kappa was 0.83 for all models. Fracture predictions were visualised with all three methods at different levels of detail, ranking from downsampled image region for classification over bounding box for detection to single pixel-level delineation for segmentation. CONCLUSIONS: All three investigated approaches reached high performances for detection of distal radius fractures with simple preprocessing and postprocessing protocols on the custom-trained models. Despite their underlying structural differences, selection of one's fracture analysis AI tool in the frame of this study reduces to the desired flavour of automation: automated classification, AI-assisted manual fracture reading or minimised false negatives.

Complemental Value of Microstructural and Macrostructural MRI in the Discrimination of Neurodegenerative Parkinson Syndromes.

PURPOSE: Various MRI-based techniques were tested for the differentiation of neurodegenerative Parkinson syndromes (NPS); the value of these techniques in direct comparison and combination is uncertain. We thus compared the diagnostic performance of macrostructural, single compartmental, and multicompartmental MRI in the differentiation of NPS. METHODS: We retrospectively included patients with NPS, including 136 Parkinson's disease (PD), 41 multiple system atrophy (MSA) and 32 progressive supranuclear palsy (PSP) and 27 healthy controls (HC). Macrostructural tissue probability values (TPV) were obtained by CAT12. The microstructure was assessed using a mesoscopic approach by diffusion tensor imaging (DTI), neurite orientation dispersion and density imaging (NODDI), and diffusion microstructure imaging (DMI). After an atlas-based read-out, a linear support vector machine (SVM) was trained on a training set (n = 196) and validated in an independent test cohort (n = 40). The diagnostic performance of the SVM was compared for different inputs individually and in combination. RESULTS: Regarding the inputs separately, we observed the best diagnostic performance for DMI. Overall, the combination of DMI and TPV performed best and correctly classified 88% of the patients. The corresponding area under the receiver operating characteristic curve was 0.87 for HC, 0.97 for PD, 1.0 for MSA, and 0.99 for PSP. CONCLUSION: We were able to demonstrate that (1) MRI parameters that approximate the microstructure provided substantial added value over conventional macrostructural imaging, (2) multicompartmental biophysically motivated models performed better than the single compartmental DTI and (3) combining macrostructural and microstructural information classified NPS and HC with satisfactory performance, thus suggesting a complementary value of both approaches.