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.

Automated detection of cephalometric landmarks using deep neural patchworks.

OBJECTIVES: This study evaluated the accuracy of deep neural patchworks (DNPs), a deep learning-based segmentation framework, for automated identification of 60 cephalometric landmarks (bone-, soft tissue- and tooth-landmarks) on CT scans. The aim was to determine whether DNP could be used for routine three-dimensional cephalometric analysis in diagnostics and treatment planning in orthognathic surgery and orthodontics. METHODS: Full skull CT scans of 30 adult patients (18 female, 12 male, mean age 35.6 years) were randomly divided into a training and test data set (each n = 15). Clinician A annotated 60 landmarks in all 30 CT scans. Clinician B annotated 60 landmarks in the test data set only. The DNP was trained using spherical segmentations of the adjacent tissue for each landmark. Automated landmark predictions in the separate test data set were created by calculating the center of mass of the predictions. The accuracy of the method was evaluated by comparing these annotations to the manual annotations. RESULTS: The DNP was successfully trained to identify all 60 landmarks. The mean error of our method was 1.94 mm (SD 1.45 mm) compared to a mean error of 1.32 mm (SD 1.08 mm) for manual annotations. The minimum error was found for landmarks ANS 1.11 mm, SN 1.2 mm, and CP_R 1.25 mm. CONCLUSION: The DNP-algorithm was able to accurately identify cephalometric landmarks with mean errors <2 mm. This method could improve the workflow of cephalometric analysis in orthodontics and orthognathic surgery. Low training requirements while still accomplishing high precision make this method particularly promising for clinical use.

Multiparametric magnetic resonance imaging for radiation therapy response monitoring in soft tissue sarcomas: a histology and MRI co-registration algorithm.

Rationale: To establish a spatially exact co-registration procedure between in vivo multiparametric magnetic resonance imaging (mpMRI) and (immuno)histopathology of soft tissue sarcomas (STS) to identify imaging parameters that reflect radiation therapy response of STS. Methods: The mpMRI-Protocol included diffusion-weighted (DWI), intravoxel-incoherent motion (IVIM), and dynamic contrast-enhancing (DCE) imaging. The resection specimen was embedded in 6.5% agarose after initial fixation in formalin. To ensure identical alignment of histopathological sectioning and in vivo imaging, an ex vivo MRI scan of the specimen was rigidly co-registered with the in vivo mpMRI. The deviating angulation of the specimen to the in vivo location of the tumor was determined. The agarose block was trimmed accordingly. A second ex vivo MRI in a dedicated localizer with a 4 mm grid was performed, which was matched to a custom-built sectioning machine. Microtomy sections were stained with hematoxylin and eosin. Immunohistochemical staining was performed with anti-ALDH1A1 antibodies as a radioresistance and anti-MIB1 antibodies as a proliferation marker. Fusion of the digitized microtomy sections with the in vivo mpMRI was accomplished through nonrigid co-registration to the in vivo mpMRI. Co-registration accuracy was qualitatively assessed by visual assessment and quantitatively evaluated by computing target registration errors (TRE). Results: The study sample comprised nine tumor sections from three STS patients. Visual assessment after nonrigid co-registration showed a strong morphological correlation of the histopathological specimens with ex vivo MRI and in vivo mpMRI after neoadjuvant radiation therapy. Quantitative assessment of the co-registration procedure using TRE analysis of different pairs of pathology and MRI sections revealed highly accurate structural alignment, with a total median TRE of 2.25 mm (histology - ex vivo MRI), 2.22 mm (histology - in vivo mpMRI), and 2.02 mm (ex vivo MRI - in vivo mpMRI). There was no significant difference between TREs of the different pairs of sections or caudal, middle, and cranial tumor parts, respectively. Conclusion: Our initial results show a promising approach to obtaining accurate co-registration between histopathology and in vivo MRI for STS. In a larger cohort of patients, the method established here will enable the prospective identification and validation of in vivo imaging biomarkers for radiation therapy response prediction and monitoring in STS patients via precise molecular and cellular correlation.

Multiclass datasets expand neural network utility: an example on ankle radiographs.

PURPOSE: Artificial intelligence in computer vision has been increasingly adapted in clinical application since the implementation of neural networks, potentially providing incremental information beyond the mere detection of pathology. As its algorithmic approach propagates input variation, neural networks could be used to identify and evaluate relevant image features. In this study, we introduce a basic dataset structure and demonstrate a pertaining use case. METHODS: A multidimensional classification of ankle x-rays (n = 1493) rating a variety of features including fracture certainty was used to confirm its usability for separating input variations. We trained a customized neural network on the task of fracture detection using a state-of-the-art preprocessing and training protocol. By grouping the radiographs into subsets according to their image features, the influence of selected features on model performance was evaluated via selective training. RESULTS: The models trained on our dataset outperformed most comparable models of current literature with an ROC AUC of 0.943. Excluding ankle x-rays with signs of surgery improved fracture classification performance (AUC 0.955), while limiting the training set to only healthy ankles with and without fracture had no consistent effect. CONCLUSION: Using multiclass datasets and comparing model performance, we were able to demonstrate signs of surgery as a confounding factor, which, following elimination, improved our model. Also eliminating pathologies other than fracture in contrast had no effect on model performance, suggesting a beneficial influence of feature variability for robust model training. Thus, multiclass datasets allow for evaluation of distinct image features, deepening our understanding of pathology imaging.

Evaluation of computed tomography settings in the context of visualization and discrimination of low dose injections of a novel liquid soft tissue fiducial marker in head and neck imaging.

BACKGROUND: Intraoperative incorporation of radiopaque fiducial markers at the tumor resection surface can provide useful assistance in identifying the tumor bed in postoperative imaging for RT planning and radiological follow-up. Besides titanium clips, iodine containing injectable liquid fiducial markers represent an option that has emerged more recently for this purpose. In this study, marking oral soft tissue resection surfaces, applying low dose injections of a novel Conformité Européenne (CE)-marked liquid fiducial marker based on sucrose acetoisobutyrate (SAIB) and iodinated SAIB (x-SAIB) was investigated. METHODS: Visibility and discriminability of low dose injections of SAIB/x-SAIB (10 µl, 20 µl, 30 µl) were systematically studied at different kV settings used in clinical routine in an ex-vivo porcine mandible model. Transferability of the preclinical results into the clinical setting and applicability of DE-CT were investigated in initial patients. RESULTS: Markers created by injection volumes as low as 10 µl were visible in CT imaging at all kV settings applied in clinical routine (70-120 kV). An injection volume of 30 µl allowed differentiation from an injection volume of 10 µl. In a total of 118 injections performed in two head and neck cancer patients, markers were clearly visible in 83% and 86% of injections. DE-CT allowed for differentiation between SAIB/x-SAIB markers and other hyperdense structures. CONCLUSIONS: Injection of low doses of SAIB/x-SAIB was found to be a feasible approach to mark oral soft tissue resection surfaces, with injection volumes as low as 10 µl found to be visible at all kV settings applied in clinical routine. With the application of SAIB/x-SAIB reported for tumors of different organs already, mostly applying relatively large volumes for IGRT, this study adds information on the applicability of low dose injections to facilitate identification of the tumor bed in postoperative CT and on performance of the marker at different kV settings used in clinical routine.

SVM-Based Normal Pressure Hydrocephalus Detection.

BACKGROUND AND PURPOSE: As magnetic resonance imaging (MRI) signs of normal pressure hydrocephalus (NPH) may precede clinical symptoms we sought to evaluate an algorithm that automatically detects this pattern. METHODS: A support vector machine (SVM) was trained in 30 NPH patients treated with ventriculoperitoneal shunts and 30 healthy controls. For comparison, four neuroradiologists visually assessed sagittal MPRAGE images and graded them as no NPH pattern, possible NPH pattern, or definite NPH pattern. RESULTS: Human accuracy to visually detect a NPH was between 0.85 and 0.97. Interobserver agreement was substantial (κ = 0.656). Accuracy of the SVM algorithm was 0.93 and AUROC 0.99. Among 272 prespecified regions, gray matter and CSF volumes of both caudate, the right parietal operculum, the left basal forebrain, and the 4th ventricle showed the highest discriminative power to separate a NPH and a no NPH pattern. CONCLUSION: A NPH pattern can be reliably detected using a support vector machine (SVM). Its role in the work-up of asymptomatic patients or neurodegenerative disease has to be evaluated.

High Stimulus-Related Information in Barrel Cortex Inhibitory Interneurons.

The manner in which populations of inhibitory (INH) and excitatory (EXC) neocortical neurons collectively encode stimulus-related information is a fundamental, yet still unresolved question. Here we address this question by simultaneously recording with large-scale multi-electrode arrays (of up to 128 channels) the activity of cell ensembles (of up to 74 neurons) distributed along all layers of 3-4 neighboring cortical columns in the anesthetized adult rat somatosensory barrel cortex in vivo. Using two different whisker stimulus modalities (location and frequency) we show that individual INH neurons--classified as such according to their distinct extracellular spike waveforms--discriminate better between restricted sets of stimuli (≤6 stimulus classes) than EXC neurons in granular and infra-granular layers. We also demonstrate that ensembles of INH cells jointly provide as much information about such stimuli as comparable ensembles containing the ~20% most informative EXC neurons, however presenting less information redundancy - a result which was consistent when applying both theoretical information measurements and linear discriminant analysis classifiers. These results suggest that a consortium of INH neurons dominates the information conveyed to the neocortical network, thereby efficiently processing incoming sensory activity. This conclusion extends our view on the role of the inhibitory system to orchestrate cortical activity.

Laminar and Columnar Structure of Sensory-Evoked Multineuronal Spike Sequences in Adult Rat Barrel Cortex In Vivo.

One of the most relevant questions regarding the function of the nervous system is how sensory information is represented in populations of cortical neurons. Despite its importance, the manner in which sensory-evoked activity propagates across neocortical layers and columns has yet not been fully characterized. In this study, we took advantage of the distinct organization of the rodent barrel cortex and recorded with multielectrode arrays simultaneously from up to 74 neurons localized in several functionally identified layers and columns of anesthetized adult Wistar rats in vivo. The flow of activity within neuronal populations was characterized by temporally precise spike sequences, which were repeatedly evoked by single-whisker stimulation. The majority of the spike sequences representing instantaneous responses were led by a subgroup of putative inhibitory neurons in the principal column at thalamo-recipient layers, thus revealing the presence of feedforward inhibition. However, later spike sequences were mainly led by infragranular excitatory neurons in neighboring columns. Although the starting point of the sequences was anatomically confined, their ending point was rather scattered, suggesting that the population responses are structurally dispersed. Our data show for the first time the simultaneous intra- and intercolumnar processing of information at high temporal resolution.

Anisotropic interfacial tension, contact angles, and line tensions: a graphics-processing-unit-based Monte Carlo study of the Ising model.

As a generic example for crystals where the crystal-fluid interface tension depends on the orientation of the interface relative to the crystal lattice axes, the nearest-neighbor Ising model on the simple cubic lattice is studied over a wide temperature range, both above and below the roughening transition temperature. Using a thin-film geometry L(x)×L(y)×L(z) with periodic boundary conditions along the z axis and two free L(x)×L(y) surfaces at which opposing surface fields ±H(1) act, under conditions of partial wetting, a single planar interface inclined under a contact angle θ<π/2 relative to the yz plane is stabilized. In the y direction, a generalization of the antiperiodic boundary condition is used that maintains the translational invariance in the y direction despite the inhomogeneity of the magnetization distribution in this system. This geometry allows a simultaneous study of the angle-dependent interface tension, the contact angle, and the line tension (which depends on the contact angle, and on temperature). All these quantities are extracted from suitable thermodynamic integration procedures. In order to keep finite-size effects as well as statistical errors small enough, rather large lattice sizes (of the order of 46 million sites) were found to be necessary, and the availability of very efficient code implementation of graphics processing units was crucial for the feasibility of this study.

Characterization of two vitellogenin cDNAs from a Pandalus shrimp (Pandalopsis japonica): expression in hepatopancreas is down-regulated by endosulfan exposure.

Endosulfan is a neurotoxic organochlorine insecticide of the cyclodiene family of pesticides that inhibits molting and reproduction in aquatic crustaceans. In order to determine the molecular mechanism of endosulfan as an endocrine disrupting chemical (EDC), differential display RT-PCR (DDRT-PCR) was used to isolate genes in the shrimp, Pandalopsis japonica, affected by endosulfan exposure. PCR screening of cDNA from the hepatopancreas from control and endosulfan-exposed animals, using 120 sets of random primers, yielded partial cDNAs encoding two vitellogenin-like proteins (Pj-Vg1 and -Vg2). Complete sequences were obtained using a combination of RT-PCR and RACE-PCR. Pj-Vg1 (7883bp) encoded a protein composed of 2533 amino acid residues (283.27 kDa estimated mass), whereas Pj-Vg2 (7792 bp) encoded a protein composed of 2537 amino acids residues (284.87 kDa estimated mass). Alignment of the Pj-Vgs with those of other vitellogenins identified a conserved subtilisin cleavage site (RQKR) and the lipoprotein N-terminal (vitellin), DUF1081, and von Willebrand factor type D domains, indicating both genes encoded functional proteins. Phylogenetic analysis showed that Pj-Vg1 and -Vg2 were most similar to Pandalus hypsinotus Vg. Both Pj-Vg1 and -Vg2 were expressed primarily in the hepatopancreas, although the Pj-Vg2 transcript was also detected in the ovary. The effects of the 3-day endosulfan exposure (2.5 microg/L and 25 microg/L) on Vg expression in the hepatopancreas were determined by quantitative RT-PCR. Expression of both transcripts was significantly inhibited at 25 microg/L suggesting that Pj-Vgs can be used as indicator for endosulfan exposure.