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.

Denoising diffusion probabilistic models for 3D medical image generation.

Recent advances in computer vision have shown promising results in image generation. Diffusion probabilistic models have generated realistic images from textual input, as demonstrated by DALL-E 2, Imagen, and Stable Diffusion. However, their use in medicine, where imaging data typically comprises three-dimensional volumes, has not been systematically evaluated. Synthetic images may play a crucial role in privacy-preserving artificial intelligence and can also be used to augment small datasets. We show that diffusion probabilistic models can synthesize high-quality medical data for magnetic resonance imaging (MRI) and computed tomography (CT). For quantitative evaluation, two radiologists rated the quality of the synthesized images regarding "realistic image appearance", "anatomical correctness", and "consistency between slices". Furthermore, we demonstrate that synthetic images can be used in self-supervised pre-training and improve the performance of breast segmentation models when data is scarce (Dice scores, 0.91 [without synthetic data], 0.95 [with synthetic data]).

r2ogs5: Calibration of Numerical Groundwater Flow Models with Bayesian Optimization in R.

Large-scale and high-resolution groundwater models are currently becoming increasingly important in order to clarify the extent to which climate trends and extreme weather affect the groundwater balance regionally. As a result, the parameterization of groundwater models is becoming more detailed and more complex, making conventional calibration methods too time-consuming. Moderating the computational demand to find optimal solutions for the resulting potentially multi-modal objective function requires intelligent and efficient global optimization methods. Moreover, the increasing use of modern scripting languages R and Python to craft environmental analysis workflows calls to integrate groundwater flow simulators in such. Here we introduce and exemplify r2ogs5, a tool that integrates version 5 of the open-source simulation software OpenGeoSys into the programming language R. r2ogs5 allows for calibration of numerical groundwater flow models with a sequential model-based optimization approach that combines Bayesian optimization (BO) with surrogate modeling. Here, we describe the structure and function of r2ogs5 as well as the implemented calibration method. We then demonstrate the calibration method by calibrating 4 and 12 parameters of two simple groundwater flow models. The results indicate that this method needs fewer runs of the groundwater flow model than conventional gradient search and Latin hypercube sampling in case of the 12 parameter model. We believe that our method offers the potential to calibrate computationally expensive groundwater flow models. r2ogs5 supports groundwater flow modelers to access the statistical analysis and visualization capabilities of the R language and is a valuable tool for geoscientists already using R.

Artificial Intelligence for Clinical Interpretation of Bedside Chest Radiographs.

Background Supine chest radiography for bedridden patients in intensive care units (ICUs) is one of the most frequently ordered imaging studies worldwide. Purpose To evaluate the diagnostic performance of a neural network-based model that is trained on structured semiquantitative radiologic reports of bedside chest radiographs. Materials and Methods For this retrospective single-center study, children and adults in the ICU of a university hospital who had been imaged using bedside chest radiography from January 2009 to December 2020 were reported by using a structured and itemized template. Ninety-eight radiologists rated the radiographs semiquantitatively for the severity of disease patterns. These data were used to train a neural network to identify cardiomegaly, pulmonary congestion, pleural effusion, pulmonary opacities, and atelectasis. A held-out internal test set (100 radiographs from 100 patients) that was assessed independently by an expert panel of six radiologists provided the ground truth. Individual assessments by each of these six radiologists, by two nonradiologist physicians in the ICU, and by the neural network were compared with the ground truth. Separately, the nonradiologist physicians assessed the images without and with preliminary readings provided by the neural network. The weighted Cohen κ coefficient was used to measure agreement between the readers and the ground truth. Results A total of 193 566 radiographs in 45 016 patients (mean age, 66 years ± 16 [SD]; 61% men) were included and divided into training (n = 122 294; 64%), validation (n = 31 243; 16%), and test (n = 40 029; 20%) sets. The neural network exhibited higher agreement with a majority vote of the expert panel (κ = 0.86) than each individual radiologist compared with the majority vote of the expert panel (κ = 0.81 to ≤0.84). When the neural network provided preliminary readings, the reports of the nonradiologist physicians improved considerably (aided vs unaided, κ = 0.87 vs 0.79, respectively; P < .001). Conclusion A neural network trained with structured semiquantitative bedside chest radiography reports allowed nonradiologist physicians improved interpretations compared with the consensus reading of expert radiologists. © RSNA, 2022 Supplemental material is available for this article. See also the editorial by Wielpütz in this issue.

Comprehensive Assessment of Medial Knee Joint Instability by Valgus Stress MRI.

Standard clinical MRI techniques provide morphologic insights into knee joint pathologies, yet do not allow evaluation of ligament functionality or joint instability. We aimed to study valgus stress MRI, combined with sophisticated image post-processing, in a graded model of medial knee joint injury. To this end, eleven human cadaveric knee joint specimens were subjected to sequential injuries to the superficial medial collateral ligament (sMCL) and the anterior cruciate ligament (ACL). Specimens were imaged in 30° of flexion in the unloaded and loaded configurations (15 kp) and in the intact, partially sMCL-deficient, completely sMCL-deficient, and sMCL- and ACL-deficient conditions using morphologic sequences and a dedicated pressure-controlled loading device. Based on manual segmentations, sophisticated 3D joint models were generated to compute subchondral cortical distances for each condition and configuration. Statistical analysis included appropriate parametric tests. The medial compartment opened gradually as a function of loading and injury, especially anteriorly. Corresponding manual reference measurements by two readers confirmed these findings. Once validated in clinical trials, valgus stress MRI may comprehensively quantify medial compartment opening as a functional imaging surrogate of medial knee joint instability and qualify as an adjunct diagnostic tool in the differential diagnosis, therapeutic decision-making, and monitoring of treatment outcomes.

In-Situ Cartilage Functionality Assessment Based on Advanced MRI Techniques and Precise Compartmental Knee Joint Loading through Varus and Valgus Stress.

Stress MRI brings together mechanical loading and MRI in the functional assessment of cartilage and meniscus, yet lacks basic scientific validation. This study assessed the response-to-loading patterns of cartilage and meniscus incurred by standardized compartmental varus and valgus loading of the human knee joint. Eight human cadaveric knee joints underwent imaging by morphologic (i.e., proton density-weighted fat-saturated and 3D water-selective) and quantitative (i.e., T1ρ and T2 mapping) sequences, both unloaded and loaded to 73.5 N, 147.1 N, and 220.6 N of compartmental pressurization. After manual segmentation of cartilage and meniscus, morphometric measures and T2 and T1ρ relaxation times were quantified. CT-based analysis of joint alignment and histologic and biomechanical tissue measures served as references. Under loading, we observed significant decreases in cartilage thickness (p < 0.001 (repeated measures ANOVA)) and T1ρ relaxation times (p = 0.001; medial meniscus, lateral tibia; (Friedman test)), significant increases in T2 relaxation times (p ≤ 0.004; medial femur, lateral tibia; (Friedman test)), and adaptive joint motion. In conclusion, varus and valgus stress MRI induces meaningful changes in cartilage and meniscus secondary to compartmental loading that may be assessed by cartilage morphometric measures as well as T2 and T1ρ mapping as imaging surrogates of tissue functionality.

Seeing Beyond Morphology-Standardized Stress MRI to Assess Human Knee Joint Instability.

While providing the reference imaging modality for joint pathologies, MRI is focused on morphology and static configurations, thereby not fully exploiting the modality's diagnostic capabilities. This study aimed to assess the diagnostic value of stress MRI combining imaging and loading in differentiating partial versus complete anterior cruciate ligament (ACL)-injury. Ten human cadaveric knee joint specimens were subjected to serial imaging using a 3.0T MRI scanner and a custom-made pressure-controlled loading device. Emulating the anterior-drawer test, joints were imaged before and after arthroscopic partial and complete ACL transection in the unloaded and loaded configurations using morphologic sequences. Following manual segmentations and registration of anatomic landmarks, two 3D vectors were computed between anatomic landmarks and registered coordinates. Loading-induced changes were quantified as vector lengths, angles, and projections on the x-, y-, and z-axis, related to the intact unloaded configuration, and referenced to manual measurements. Vector lengths and projections significantly increased with loading and increasing ACL injury and indicated multidimensional changes. Manual measurements confirmed gradually increasing anterior tibial translation. Beyond imaging of ligament structure and functionality, stress MRI techniques can quantify joint stability to differentiate partial and complete ACL injury and, possibly, compare surgical procedures and monitor treatment outcomes.

Identifying the imaging correlates of cartilage functionality based on quantitative MRI mapping - The collagenase exposure model.

Cartilage functionality is determined by tissue structure and composition. If altered, cartilage is predisposed to premature degeneration. This pathomimetical study of early osteoarthritis evaluated the dose-dependant effects of collagenase-induced collagen disintegration and proteoglycan depletion on cartilage functionality as assessed by serial T1, T1ρ, T2, and T2* mapping under loading. 30 human femoral osteochondral samples underwent imaging on a clinical 3.0 T MRI scanner (Achieva, Philips) in the unloaded reference configuration (δ0) and under pressure-controlled quasi-static indentation loading to 15.1 N (δ1) and to 28.6 N (δ2). Imaging was performed before and after exposure to low (LC, 0.5 mg/mL; n = 10) or high concentration (HC, 1.5 mg/mL; n = 10) of collagenase. Untreated samples served as controls (n = 10). Loading responses were determined for the entire sample and the directly loaded (i.e. sub-pistonal) and bilaterally adjacent (i.e. peri­pistonal) regions, referenced histologically, quantified as relative changes, and analysed using adequate parametric and non-parametric statistical tests. Dose-dependant surface disintegration and tissue loss were reflected by distinctly different pre- and post-exposure response-to-loading patterns. While T1 generally decreased with loading, regardless of collagenase exposure, T1ρ increased significantly after HC exposure (p = 0.008). Loading-induced decreases in T2 were significant after LC exposure (p = 0.006), while changes in T2* were ambiguous. In conclusion, aberrant loading-induced changes in T2 and T1ρ reflect moderate and severe matrix changes, respectively, and indicate the close interrelatedness of matrix changes and functionality in cartilage.

Magnetic resonance imaging of human knee joint functionality under variable compressive in-situ loading and axis alignment.

Magnetic resonance imaging (MRI) under mechanical loading, commonly referred to as stress MRI, allows the evaluation of functional properties of intra- and periarticular tissues non-invasively beyond static assessment. Quantitative MRI can identify physiological and pathological responses to loading as indication of, potentially treatable, early degeneration and load transmission failure. Therefore, we have developed and validated an MRI-compatible pressure-controlled axial loading device to compress human knee specimens under variable loading intensity and axis deviation. Ten structurally intact human knee specimens (mean age 83.2 years) were studied on a 3.0T scanner (Achieva, Philips). Proton density-weighted fat-saturated turbo spin-echo and high-resolution 3D water selective 3D gradient-echo MRI scans were acquired sequentially at 10° joint flexion in seven configurations: unloaded and then at approximately half and full body weight loading in neutral, 10° varus and 10° valgus alignment, respectively. Following manual segmentation in both femorotibial compartments, cartilage thickness (ThC) was determined as well as meniscus extrusion (ExM). These measures were compared to computed tomography scans, histological grading (Mankin and Pauli scores), and biomechanical properties (Instantaneous Young's Modulus). Compartmental, regional and subregional changes in ThC and ExM were reflective of loading intensity and joint alignment, with the greatest changes observed in the medial compartment during varus and in the lateral compartment during valgus loading. These were not significantly associated with the histological tissue status or biomechanical properties. In conclusion, this study explores the physiological in-situ response of knee cartilage and meniscus, based on stress MRI, and as a function of loading intensity, joint alignment, histological tissue status, and biomechanical properties, as another step towards clinical implementation.

An MRI-compatible varus-valgus loading device for whole-knee joint functionality assessment based on compartmental compression: a proof-of-concept study.

OBJECTIVE: Beyond static assessment, functional techniques are increasingly applied in magnetic resonance imaging (MRI) studies. Stress MRI techniques bring together MRI and mechanical loading to study knee joint and tissue functionality, yet prototypical axial compressive loading devices are bulky and complex to operate. This study aimed to design and validate an MRI-compatible pressure-controlled varus-valgus loading device that applies loading along the joint line. METHODS: Following the device's thorough validation, we demonstrated proof of concept by subjecting a structurally intact human cadaveric knee joint to serial imaging in unloaded and loaded configurations, i.e. to varus and valgus loading at 7.5 kPa (= 73.5 N), 15 kPa (= 147.1 N), and 22.5 kPa (= 220.6 N). Following clinical standard (PDw fs) and high-resolution 3D water-selective cartilage (WATSc) sequences, we performed manual segmentations and computations of morphometric cartilage measures. We used CT and radiography (to quantify joint space widths) and histology and biomechanics (to assess tissue quality) as references. RESULTS: We found (sub)regional decreases in cartilage volume, thickness, and mean joint space widths reflective of areal pressurization of the medial and lateral femorotibial compartments. DISCUSSION: Once substantiated by larger sample sizes, varus-valgus loading may provide a powerful alternative stress MRI technique.

Multiphase CT-based prediction of Child-Pugh classification: a machine learning approach.

BACKGROUND: To evaluate whether machine learning algorithms allow the prediction of Child-Pugh classification on clinical multiphase computed tomography (CT). METHODS: A total of 259 patients who underwent diagnostic abdominal CT (unenhanced, contrast-enhanced arterial, and venous phases) were included in this retrospective study. Child-Pugh scores were determined based on laboratory and clinical parameters. Linear regression (LR), Random Forest (RF), and convolutional neural network (CNN) algorithms were used to predict the Child-Pugh class. Their performances were compared to the prediction of experienced radiologists (ERs). Spearman correlation coefficients and accuracy were assessed for all predictive models. Additionally, a binary classification in low disease severity (Child-Pugh class A) and advanced disease severity (Child-Pugh class ≥ B) was performed. RESULTS: Eleven imaging features exhibited a significant correlation when adjusted for multiple comparisons with Child-Pugh class. Significant correlations between predicted and measured Child-Pugh classes were observed (ρLA = 0.35, ρRF = 0.32, ρCNN = 0.51, ρERs = 0.60; p < 0.001). Significantly better accuracies for the prediction of Child-Pugh classes versus no-information rate were found for CNN and ERs (p ≤ 0.034), not for LR and RF (p ≥ 0.384). For binary severity classification, the area under the curve at receiver operating characteristic analysis was significantly lower (p ≤ 0.042) for LR (0.71) and RF (0.69) than for CNN (0.80) and ERs (0.76), without significant differences between CNN and ERs (p = 0.144). CONCLUSIONS: The performance of a CNN in assessing Child-Pugh class based on multiphase abdominal CT images is comparable to that of ERs.