Checklist for Artificial Intelligence in Medical Imaging Reporting Adherence in Peer-Reviewed and Preprint Manuscripts With the Highest Altmetric Attention Scores: A Meta-Research Study.

Purpose: To establish reporting adherence to the Checklist for Artificial Intelligence in Medical Imaging (CLAIM) in diagnostic accuracy AI studies with the highest Altmetric Attention Scores (AAS), and to compare completeness of reporting between peer-reviewed manuscripts and preprints. Methods: MEDLINE, EMBASE, arXiv, bioRxiv, and medRxiv were retrospectively searched for 100 diagnostic accuracy medical imaging AI studies in peer-reviewed journals and preprint platforms with the highest AAS since the release of CLAIM to June 24, 2021. Studies were evaluated for adherence to the 42-item CLAIM checklist with comparison between peer-reviewed manuscripts and preprints. The impact of additional factors was explored including body region, models on COVID-19 diagnosis and journal impact factor. Results: Median CLAIM adherence was 48% (20/42). The median CLAIM score of manuscripts published in peer-reviewed journals was higher than preprints, 57% (24/42) vs 40% (16/42), P < .0001. Chest radiology was the body region with the least complete reporting (P = .0352), with manuscripts on COVID-19 less complete than others (43% vs 54%, P = .0002). For studies published in peer-reviewed journals with an impact factor, the CLAIM score correlated with impact factor, rho = 0.43, P = .0040. Completeness of reporting based on CLAIM score had a positive correlation with a study's AAS, rho = 0.68, P < .0001. Conclusions: Overall reporting adherence to CLAIM is low in imaging diagnostic accuracy AI studies with the highest AAS, with preprints reporting fewer study details than peer-reviewed manuscripts. Improved CLAIM adherence could promote adoption of AI into clinical practice and facilitate investigators building upon prior works.

Brain simulation augments machine-learning-based classification of dementia.

Introduction: Computational brain network modeling using The Virtual Brain (TVB) simulation platform acts synergistically with machine learning (ML) and multi-modal neuroimaging to reveal mechanisms and improve diagnostics in Alzheimer's disease (AD). Methods: We enhance large-scale whole-brain simulation in TVB with a cause-and-effect model linking local amyloid beta (Aß) positron emission tomography (PET) with altered excitability. We use PET and magnetic resonance imaging (MRI) data from 33 participants of the Alzheimer's Disease Neuroimaging Initiative (ADNI3) combined with frequency compositions of TVB-simulated local field potentials (LFP) for ML classification. Results: The combination of empirical neuroimaging features and simulated LFPs significantly outperformed the classification accuracy of empirical data alone by about 10% (weighted F1-score empirical 64.34% vs. combined 74.28%). Informative features showed high biological plausibility regarding the AD-typical spatial distribution. Discussion: The cause-and-effect implementation of local hyperexcitation caused by Aß can improve the ML-driven classification of AD and demonstrates TVB's ability to decode information in empirical data using connectivity-based brain simulation.

Quantitative and qualitative liver CT: imaging feature association with histopathologically confirmed hepatic cirrhosis.

PURPOSE: To assess the diagnostic performance of quantitative and qualitative imaging features of hepatic cirrhosis on CT. METHODS: A single-center retrospective cohort study was performed on all patients who had undergone non-targeted liver biopsy < 3 months following abdominal CT imaging between 2007 and 2020. Histopathology was required as a reference standard for hepatic cirrhosis diagnosis. Two readers independently assessed all CT quantitative and qualitative features, blinded to the clinical history and the reference standard. The diagnostic performance of each imaging feature was assessed using multivariate regression and logistic regression in a recursive feature elimination framework. RESULTS: 98 consecutive patients met inclusion criteria including 26 with histopathologically confirmed hepatic cirrhosis, and 72 without cirrhosis. Liver surface nodularity (p < 0.0001), lobar redistribution (p < 0.0001), and expanded gallbladder fossa (p < 0.0016) were qualitative CT features associated with liver cirrhosis consistent between both reviewers. Liver surface nodularity demonstrated highest sensitivity (73-77%) and specificity (79-82%). Falciform space width was the only quantitative feature associated with cirrhosis, for a single reviewer (p < 0.04). Using a recursive feature elimination framework, liver surface nodularity and falciform space width were the strongest performing features for identifying cirrhosis. No feature combinations strengthened diagnostic performance. CONCLUSION: Many quantitative and qualitative CT imaging signs of hepatic cirrhosis have either poor accuracy or poor inter-observer agreement. Qualitative imaging features of hepatic cirrhosis on CT performed better than quantitative metrics, with liver surface nodularity the most optimal feature for diagnosing hepatic cirrhosis.

Radiologist Productivity Analytics: Factors Impacting Abdominal Pelvic CT Exam Reporting Times.

The purpose is to determine factors impacting radiologist abdominal pelvic CT exam reporting time. This study was Research Ethics Board approved. Between January 2019 and March 2020, consecutive abdominal pelvic CT exams were documented as structured or unstructured based on application of templates with separate sections for different organs or organ systems. Radiologist reporting location, patient class (inpatient, Emergency Department (ED) patient, outpatient), radiologist fellowship-training, report word count, and radiologist years of experience were documented. Median reporting times were compared using the Wilcoxon Rank-sum test, Kruskal-Wallis test, and regression analysis. Spearman's rank correlation was used to determine correlation between word count and radiologist experience with reporting time. P < 0.05 is defined statistical significance. A total of 3602 abdominal pelvic CT exam reports completed by 33 radiologists were reviewed, including 1150 outpatient and 2452 inpatient and Emergency Department (ED) cases. 1398 of all reports were structured. Median reporting time for structured and unstructured reports did not differ (P = 0.870). Reports dictated in-house were completed faster than reports dictated remotely (P < 0.001), and reports for inpatients/ED patients were completed faster than for outpatients (P < 0.001). Reporting time differences existed between radiologists (P < 0.001) that were not explained by fellowship training (P = 0.762). Median reporting time had a weak correlation with word count (ρ = 0.355) and almost no correlation with radiologist years of experience (ρ = 0.167), P < 0.001. Abdominal pelvic CT reporting is most efficient when dictations are completed in-house and for high-priority cases; the use of structured templates, radiologist fellowship training, and years of experience have no impact on reporting times.

MRI LI-RADS Version 2018: Impact of and Reduction in Ancillary Features.

OBJECTIVE. The purpose of this study is to determine the impact of LI-RADS ancillary features on MRI and to ascertain whether the number of ancillary features can be reduced without compromising LI-RADS accuracy. MATERIALS AND METHODS. A total of 222 liver observations in 81 consecutive patients were identified on MRI between August 2013 and December 2018. The presence or absence of major and ancillary features was used to determine the LI-RADS category for LR-1 to LR-5 observations. Final diagnosis was established on the basis of pathologic findings or one of several composite clinical reference standards. Diagnostic accuracy was compared with and without ancillary features by use of the z test of proportions. Decision tree analysis and machine learning-based feature pruning were used to identify noncontributory ancillary features for LI-RADS categorization. Interobserver agreement with and without ancillary features was measured using the Krippendorff alpha coefficient, and comparisons were made using bootstrapping. A p < .05 was considered statistically significant. RESULTS. Application of ancillary features resulted in a change in the LI-RADS category of seven hepatocellular carcinomas (HCCs), with the category of six of seven (86%) HCCs upgraded; 51 benign observations also had a change in LI-RADS category, with the category of 33 (65%) of these observations downgraded. When ancillary features were applied, the percentage of HCCs in each LI-RADS category did not differ significantly compared with major features alone (p = .06-.49). Decision tree analysis and the machine learning model identified five ancillary features as noncontributory: corona enhancement, nodule-in-nodule, mosaic architecture, blood products in mass, and fat in a mass, more than in adjacent liver. Interobserver agreement was high with and without application of ancillary features; however, it was significantly higher without ancillary features (p < .001). CONCLUSION. Although ancillary features are an important component of LI-RADS, their impact may be small. Several ancillary features likely can be removed from LI-RADS without compromising diagnostic performance.

Pancreas adenocarcinoma CT texture analysis: comparison of 3D and 2D tumor segmentation techniques.

PURPOSE: To determine equivalency of multi-slice 3D CTTA and single slice 2D CTTA of pancreas adenocarcinoma. METHODS: This retrospective study was research ethics board approved. Untreated pancreas adenocarcinomas were segmented on CT in 128 consecutive patients. Tumor segmentation was compared using two techniques: 3D segmentation by contouring all visible tumor in a 3D volume, and 2D segmentation using only a single axial image. First-order CTTA features including mean, minimum, maximum Hounsfield units (HU), standard deviation, skewness, kurtosis, entropy, and second-order gray-level co-occurrence matrix (GLCM) features homogeneity, contrast, correlation, entropy and dissimilarity were extracted. Median values were compared using the Mann-Whitney U test with Holm-Bonferroni correction. Kendall's Rank Correlation Tau assessed for correlation, and agreement was calculated using intraclass correlation coefficients (ICC) using a two-way model with single rating and absolute agreement. Statistical significance defined as P < 0.05. RESULTS: The median values of CTTA features differed significantly between 3 and 2D segmentations for all of the evaluated features except for mean attenuation, standard deviation and skewness (P = 0.2979 each). 3D and 2D segmentations had moderate correlation for mean attenuation (R = 0.69, P < 0.01), while all other features demonstrated poor to fair correlation. Agreement between 3 and 2D segmentations was good for mean attenuation (ICC: 0.87, P < 0.01), moderate for minimum (ICC: 0.65, P < 0.01) and standard deviation (ICC: 0.56, P < 0.01), and poor for all other features. CONCLUSION: While pancreas adenocarcinoma CTTA features obtained using 3D and 2D segmentation have multiple associations with clinically relevant outcomes, these segmentation techniques are likely not interchangeable other than for mean HU.

On the time-course of functional connectivity: theory of a dynamic progression of concussion effects.

The current literature presents a discordant view of mild traumatic brain injury and its effects on the human brain. This dissonance has often been attributed to heterogeneities in study populations, aetiology, acuteness, experimental paradigms and/or testing modalities. To investigate the progression of mild traumatic brain injury in the human brain, the present study employed data from 93 subjects (48 healthy controls) representing both acute and chronic stages of mild traumatic brain injury. The effects of concussion across different stages of injury were measured using two metrics of functional connectivity in segments of electroencephalography time-locked to an active oddball task. Coherence and weighted phase-lag index were calculated separately for individual frequency bands (delta, theta, alpha and beta) to measure the functional connectivity between six electrode clusters distributed from frontal to parietal regions across both hemispheres. Results show an increase in functional connectivity in the acute stage after mild traumatic brain injury, contrasted with significantly reduced functional connectivity in chronic stages of injury. This finding indicates a non-linear time-dependent effect of injury. To understand this pattern of changing functional connectivity in relation to prior evidence, we propose a new model of the time-course of the effects of mild traumatic brain injury on the brain that brings together research from multiple neuroimaging modalities and unifies the various lines of evidence that at first appear to be in conflict.

Using Deep Learning Algorithms to Grade Hydronephrosis Severity: Toward a Clinical Adjunct.

Grading hydronephrosis severity relies on subjective interpretation of renal ultrasound images. Deep learning is a data-driven algorithmic approach to classifying data, including images, presenting a promising option for grading hydronephrosis. The current study explored the potential of deep convolutional neural networks (CNN), a type of deep learning algorithm, to grade hydronephrosis ultrasound images according to the 5-point Society for Fetal Urology (SFU) classification system, and discusses its potential applications in developing decision and teaching aids for clinical practice. We developed a five-layer CNN to grade 2,420 sagittal hydronephrosis ultrasound images [191 SFU 0 (8%), 407 SFU I (17%), 666 SFU II (28%), 833 SFU III (34%), and 323 SFU IV (13%)], from 673 patients ranging from 0 to 116.29 months old (M age = 16.53, SD = 17.80). Five-way (all grades) and two-way classification problems [i.e., II vs. III, and low (0-II) vs. high (III-IV)] were explored. The CNN classified 94% (95% CI, 93-95%) of the images correctly or within one grade of the provided label in the five-way classification problem. Fifty-one percent of these images (95% CI, 49-53%) were correctly predicted, with an average weighted F1 score of 0.49 (95% CI, 0.47-0.51). The CNN achieved an average accuracy of 78% (95% CI, 75-82%) with an average weighted F1 of 0.78 (95% CI, 0.74-0.82) when classifying low vs. high grades, and an average accuracy of 71% (95% CI, 68-74%) with an average weighted F1 score of 0.71 (95% CI, 0.68-0.75) when discriminating between grades II vs. III. Our model performs well above chance level, and classifies almost all images either correctly or within one grade of the provided label. We have demonstrated the applicability of a CNN approach to hydronephrosis ultrasound image classification. Further investigation into a deep learning-based clinical adjunct for hydronephrosis is warranted.

Capturing the Forest but Missing the Trees: Microstates Inadequate for Characterizing Shorter-Scale EEG Dynamics.

The brain is known to be active even when not performing any overt cognitive tasks, and often it engages in involuntary mind wandering. This resting state has been extensively characterized in terms of fMRI-derived brain networks. However, an alternate method has recently gained popularity: EEG microstate analysis. Proponents of microstates postulate that the brain discontinuously switches between four quasi-stable states defined by specific EEG scalp topologies at peaks in the global field potential (GFP). These microstates are thought to be "atoms of thought," involved with visual, auditory, salience, and attention processing. However, this method makes some major assumptions by excluding EEG data outside the GFP peaks and then clustering the EEG scalp topologies at the GFP peaks, assuming that only one microstate is active at any given time. This study explores the evidence surrounding these assumptions by studying the temporal dynamics of microstates and its clustering space using tools from dynamical systems analysis, fractal, and chaos theory to highlight the shortcomings in microstate analysis. The results show evidence of complex and chaotic EEG dynamics outside the GFP peaks, which is being missed by microstate analysis. Furthermore, the winner-takes-all approach of only one microstate being active at a time is found to be inadequate since the dynamic EEG scalp topology does not always resemble that of the assigned microstate, and there is competition among the different microstate classes. Finally, clustering space analysis shows that the four microstates do not cluster into four distinct and separable clusters. Taken collectively, these results show that the discontinuous description of EEG microstates is inadequate when looking at nonstationary short-scale EEG dynamics.

Individualized pattern recognition for detecting mind wandering from EEG during live lectures.

NEURAL CORRELATES OF MIND WANDERING: The ability to detect mind wandering as it occurs is an important step towards improving our understanding of this phenomenon and studying its effects on learning and performance. Current detection methods typically rely on observable behaviour in laboratory settings, which do not capture the underlying neural processes and may not translate well into real-world settings. We address both of these issues by recording electroencephalography (EEG) simultaneously from 15 participants during live lectures on research in orthopedic surgery. We performed traditional group-level analysis and found neural correlates of mind wandering during live lectures that are similar to those found in some laboratory studies, including a decrease in occipitoparietal alpha power and frontal, temporal, and occipital beta power. However, individual-level analysis of these same data revealed that patterns of brain activity associated with mind wandering were more broadly distributed and highly individualized than revealed in the group-level analysis. MIND WANDERING DETECTION: To apply these findings to mind wandering detection, we used a data-driven method known as common spatial patterns to discover scalp topologies for each individual that reflects their differences in brain activity when mind wandering versus attending to lectures. This approach avoids reliance on known neural correlates primarily established through group-level statistics. Using this method for individual-level machine learning of mind wandering from EEG, we were able to achieve an average detection accuracy of 80-83%. CONCLUSIONS: Modelling mind wandering at the individual level may reveal important details about its neural correlates that are not reflected when using traditional observational and statistical methods. Using machine learning techniques for this purpose can provide new insight into the varieties of neural activity involved in mind wandering, while also enabling real-time detection of mind wandering in naturalistic settings.

From Group-Level Statistics to Single-Subject Prediction: Machine Learning Detection of Concussion in Retired Athletes.

There has been increased effort to understand the neurophysiological effects of concussion aimed to move diagnosis and identification beyond current subjective behavioral assessments that suffer from poor sensitivity. Recent evidence suggests that event-related potentials (ERPs) measured with electroencephalography (EEG) are persistent neurophysiological markers of past concussions. However, as such evidence is limited to group-level analyzes, the extent to which they enable concussion detection at the individual-level is unclear. One promising avenue of research is the use of machine learning to create quantitative predictive models that can detect prior concussions in individuals. In this paper, we translate the recent group-level findings from ERP studies of concussed individuals into a machine learning framework for performing single-subject prediction of past concussion. We found that a combination of statistics of single-subject ERPs and wavelet features yielded a classification accuracy of 81% with a sensitivity of 82% and a specificity of 80%, improving on current practice. Notably, the model was able to detect concussion effects in individuals who sustained their last injury as much as 30 years earlier. However, failure to detect past concussions in a subset of individuals suggests that the clear effects found in group-level analyses may not provide us with a full picture of the neurophysiological effects of concussion.

Progressive Thresholding: Shaping and Specificity in Automated Neurofeedback Training.

Neurofeedback has long been proposed as a promising form of adjunctive non-pharmaceutical treatment for a variety of neuropsychological disorders. However, there is much debate over its efficacy and specificity. Many suggest that specificity can only be achieved when a specially trained clinician manually updates reward thresholds that indicate to the trainee when they are modulating their brain activity correctly, during training. We present a novel fully automated reward thresholding algorithm called progressive thresholding and test it with a frontal alpha asymmetry neurofeedback protocol. Progressive thresholding uses dynamic difficulty tuning and individual-specific progress models to simulate the shaping a clinician might perform when setting reward thresholds manually. We demonstrate in a double-blind comparison that progressive thresholding leads to significantly better learning outcomes compared with current automatic reward thresholding algorithms.

Toward an Open-Ended BCI: A User-Centered Coadaptive Design.

Brain-computer interfaces (BCIs) allow users to control a device by interpreting their brain activity. For simplicity, these devices are designed to be operated by purposefully modulating specific predetermined neurophysiological signals, such as the sensorimotor rhythm. However, the ability to modulate a given neurophysiological signal is highly variable across individuals, contributing to the inconsistent performance of BCIs for different users. These differences suggest that individuals who experience poor BCI performance with one class of brain signals might have good results with another. In order to take advantage of individual abilities as they relate to BCI control, we need to move beyond the current approaches. In this letter, we explore a new BCI design aimed at a more individualized and user-focused experience, which we call open-ended BCI. Individual users were given the freedom to discover their own mental strategies as opposed to being trained to modulate a given brain signal. They then underwent multiple coadaptive training sessions with the BCI. Our first open-ended BCI performed similarly to comparable BCIs while accommodating a wider variety of mental strategies without a priori knowledge of the specific brain signals any individual might use. Post hoc analysis revealed individual differences in terms of which sensory modality yielded optimal performance. We found a large and significant effect of individual differences in background training and expertise, such as in musical training, on BCI performance. Future research should be focused on finding more generalized solutions to user training and brain state decoding methods to fully utilize the abilities of different individuals in an open-ended BCI. Accounting for each individual's areas of expertise could have important implications on BCI training and BCI application design.

Performance of case-control rare copy number variation annotation in classification of autism.

BACKGROUND: A substantial proportion of Autism Spectrum Disorder (ASD) risk resides in de novo germline and rare inherited genetic variation. In particular, rare copy number variation (CNV) contributes to ASD risk in up to 10% of ASD subjects. Despite the striking degree of genetic heterogeneity, case-control studies have detected specific burden of rare disruptive CNV for neuronal and neurodevelopmental pathways. Here, we used machine learning methods to classify ASD subjects and controls, based on rare CNV data and comprehensive gene annotations. We investigated performance of different methods and estimated the percentage of ASD subjects that could be reliably classified based on presumed etiologic CNV they carry. RESULTS: We analyzed 1,892 Caucasian ASD subjects and 2,342 matched controls. Rare CNVs (frequency 1% or less) were detected using Illumina 1M and 1M-Duo BeadChips. Conditional Inference Forest (CF) typically performed as well as or better than other classification methods. We found a maximum AUC (area under the ROC curve) of 0.533 when considering all ASD subjects with rare genic CNVs, corresponding to 7.9% correctly classified ASD subjects and less than 3% incorrectly classified controls; performance was significantly higher when considering only subjects harboring de novo or pathogenic CNVs. We also found rare losses to be more predictive than gains and that curated neurally-relevant annotations (brain expression, synaptic components and neurodevelopmental phenotypes) outperform Gene Ontology and pathway-based annotations. CONCLUSIONS: CF is an optimal classification approach for case-control rare CNV data and it can be used to prioritize subjects with variants potentially contributing to ASD risk not yet recognized. The neurally-relevant annotations used in this study could be successfully applied to rare CNV case-control data-sets for other neuropsychiatric disorders.

Examining the role of the temporo-parietal network in memory, imagery, and viewpoint transformations.

The traditional view of the medial temporal lobe (MTL) focuses on its role in episodic memory. However, some of the underlying functions of the MTL can be ascertained from its wider role in supporting spatial cognition in concert with parietal and prefrontal regions. The MTL is strongly implicated in the formation of enduring allocentric representations (e.g., O'Keefe, 1976; King et al., 2002; Ekstrom et al., 2003). According to our BBB model (Byrne et al., 2007), these representations must interact with head-centered and body-centered representations in posterior parietal cortex via a transformation circuit involving retrosplenial areas. Egocentric sensory representations in parietal areas can then cue the recall of allocentric spatial representations in long-term memory and, conversely, the products of retrieval in MTL can generate mental imagery within a parietal "window." Such imagery is necessarily egocentric and forms part of visuospatial working memory, in which it can be manipulated for the purpose of planning/imagining the future. Recent fMRI evidence (Lambrey et al., 2012; Zhang et al., 2012) supports the BBB model. To further test the model, we had participants learn the locations of objects in a virtual scene and tested their spatial memory under conditions that impose varying demands on the transformation circuit. We analyzed how brain activity correlated with accuracy in judging the direction of an object (1) from visuospatial working memory (we assume transient working memory due to the order of tasks and the absence of change in viewpoint, but long-term memory retrieval is also possible), (2) after a rotation of viewpoint, or (3) after a rotation and translation of viewpoint (judgment of relative direction). We found performance-related activity in both tasks requiring viewpoint rotation (ROT and JRD, i.e., conditions 2 and 3) in the core medial temporal to medial parietal circuit identified by the BBB model. These results are consistent with the predictions of the BBB model, and shed further light on the neural mechanisms underlying spatial memory, mental imagery and viewpoint transformations.