The Algorithmic Audit: Working with Vendors to Validate Radiology-AI Algorithms-How We Do It.

There is a plethora of Artificial Intelligence (AI) tools that are being developed around the world aiming at either speeding up or improving the accuracy of radiologists. It is essential for radiologists to work with the developers of such algorithms to determine true clinical utility and risks associated with these algorithms. We present a framework, called an Algorithmic Audit, for working with the developers of such algorithms to test and improve the performance of the algorithms. The framework includes concepts of true independent validation on data that the algorithm has not seen before, curating datasets for such testing, deep examination of false positives and false negatives (to examine implications of such errors) and real-world deployment and testing of algorithms.

Unboxing AI - Radiological Insights Into a Deep Neural Network for Lung Nodule Characterization.

RATIONALE AND OBJECTIVES: To explain predictions of a deep residual convolutional network for characterization of lung nodule by analyzing heat maps. MATERIALS AND METHODS: A 20-layer deep residual CNN was trained on 1245 Chest CTs from National Lung Screening Trial (NLST) trial to predict the malignancy risk of a nodule. We used occlusion to systematically block regions of a nodule and map drops in malignancy risk score to generate clinical attribution heatmaps on 103 nodules from Lung Image Database Consortium image collection and Image Database Resource Initiative (LIDC-IDRI) dataset, which were analyzed by a thoracic radiologist. The features were described as heat inside nodule -bright areas inside nodule, peripheral heat continuous/interrupted bright areas along nodule contours, heat in adjacent plane -brightness in scan planes juxtaposed with the nodule, satellite heat - a smaller bright spot in proximity to nodule in the same scan plane, heat map larger than nodule bright areas corresponding to the shape of the nodule seen outside the nodule margins and heat in calcification. RESULTS: These six features were assigned binary values. This feature vector was fedinto a standard J48 decision tree with 10-fold cross-validation, which gave an 85 % weighted classification accuracy with a 77.8% True Positive (TP) rate, 8% False Positive (FP) rate for benign cases and 91.8% TP and 22.2% FP rates for malignant cases. Heat Inside nodule was more frequently observed in nodules classified as malignant whereas peripheral heat, heat in adjacent plane, and satellite heat were more commonly seen in nodules classified as benign. CONCLUSION: We discuss the potential ability of a radiologist to visually parse the deep learning algorithm generated "heat map" to identify features aiding classification.

Preliminary evidence from a prospective DTI study suggests a posterior-to-anterior pattern of recovery in college athletes with sports-related concussion.

OBJECTIVES: We compared the integrity of white matter (WM) microstructure to the course of recovery in athletes who sustained one sports-related concussion (SRC), assessing individual longitudinal changes in WM fiber tracts following SRC using pre- and post-injury measurements. MATERIALS AND METHODS: Baseline diffusion tensor imaging (DTI) scans and neuropsychological tests were collected on 53 varsity contact-sport college athletes. Participants (n = 13) who subsequently sustained an SRC underwent DTI scans and neuropsychological testing at 2 days, 2 weeks, and 2 months following injury. RESULTS: Relying on tract-based spatial statistics (TBSS) analyses, we found that radial diffusivity (RD) and mean diffusivity (MD) were significantly increased at 2 days post-injury compared to the same-subject baseline (corrected p < 0.02). These alterations were visible in anterior/posterior WM regions spanning both hemispheres, demonstrating a diffuse pattern of injury after concussion. Implicated WM fiber tracts at 2 days include the following: right superior/inferior longitudinal fasciculus; right/left inferior fronto-occipital fasciculus; right corticospinal tract; right acoustic radiation; right/left anterior thalamic radiations; right/left uncinate fasciculus; and forceps major/minor. At 2 weeks post-injury, persistently elevated RD and MD were observed solely in prefrontal portions of WM fiber tracts (using same-subject contrasts). No significant differences were found for FA in any of the post-injury comparisons to baseline. Plots of individual subject RD and MD in prefrontal WM demonstrated homogenous increases from baseline to just after SRC; thereafter, trajectories became more variable. Most subjects' diffusivity values remained elevated at 2 months post-injury relative to their own baseline. Over the 2-month period after SRC, recovery of WM fiber tracts appeared to follow a posterior-to-anterior trend, paralleling the posterior-anterior pattern of WM maturation previously identified in the normal population. CONCLUSION: These results suggest greater vulnerability of prefrontal regions to SRC, underline the importance of an individualized approach to concussion management, and show promise for using RD and MD for imaging-based diagnosis of SRC.

Prospective clinical assessment using Sideline Concussion Assessment Tool-2 testing in the evaluation of sport-related concussion in college athletes.

OBJECTIVE: To evaluate the utility of the Sideline Concussion Assessment Tool (SCAT)-2 in collegiate athletes with sport-related concussion. DESIGN: Prospective cross-sectional study with baseline testing and serial repeat testing after concussion in contact sport athletes and non-concussed control athletes. SETTING: Division I University. PARTICIPANTS: Male and female club rugby and varsity athletes. INTERVENTIONS: Baseline measures of concussion symptoms, cognitive function, and balance were obtained using the SCAT-2. Serial postinjury testing was conducted as clinically indicated. MAIN OUTCOME MEASURES: The SCAT-2 total and subset scores were calculated and evaluated at baseline and after injury. RESULTS: The total SCAT-2 score and the composite scores of symptoms, symptom severity, and balance were significantly different in concussed groups after injury when compared with baseline. When comparing performance in concussed versus control athletes, all subcomponents of the SCAT-2 were significantly different. No differences in baseline SCAT-2 scores were seen based on self-reported history of concussion. At baseline, anxiety and depression screening scores were associated with higher symptom scores. When compared with baseline, a 3.5-point drop in SCAT-2 score had 96% sensitivity and 81% specificity in detecting concussion. When examined to exclude baseline scores, a cutoff value of 74.5 was associated with 83% sensitivity and 91% specificity in predicting concussion versus control status. CONCLUSIONS: The SCAT-2 total composite score and each subcomponent are useful in the assessment of concussion. As SCAT-3 is similar to SCAT-2, it is expected that it too will be a useful tool.

A longitudinal diffusion tensor imaging study assessing white matter fiber tracts after sports-related concussion.

The extent of structural injury in sports-related concussion (SRC) is central to the course of recovery, long-term effects, and the decision to return to play. In the present longitudinal study, we used diffusion tensor imaging (DTI) to assess white matter (WM) fiber tract integrity within 2 days, 2 weeks, and 2 months of concussive injury. Participants were right-handed male varsity contact-sport athletes (20.2±1.0 years of age) with a medically diagnosed SRC (no loss of consciousness). They were compared to right-handed male varsity non-contact-sport athletes serving as controls (19.9±1.7 years). We found significantly increased radial diffusivity (RD) in concussed athletes (n=12; paired t-test, tract-based spatial statistics; p<0.025) at 2 days, when compared to the 2-week postinjury time point. The increase was found in a cluster of right hemisphere voxels, spanning the posterior limb of the internal capsule (IC), the retrolenticular part of the IC, the inferior longitudinal fasciculus, the inferior fronto-occipital fasciculus (sagittal stratum), and the anterior thalamic radiation. Post-hoc, univariate, between-group (controls vs. concussed), mixed-effects analysis of the cluster showed significantly higher RD at 2 days (p=0.002), as compared to the controls, with a trend in the same direction at 2 months (p=0.11). Results for fractional anisotropy (FA) in the same cluster showed a similar, but inverted, pattern; FA was decreased at 2 days and at 2 months postinjury, when compared to healthy controls. At 2 weeks postinjury, no statistical differences between concussed and control athletes were found with regard to either RD or FA. These results support the hypothesis of increased RD and reduced FA within 72 h postinjury, followed by recovery that may extend beyond 2 weeks. RD appears to be a sensitive measure of concussive injury.

Persistent differences in patterns of brain activation after sports-related concussion: a longitudinal functional magnetic resonance imaging study.

Avoiding recurrent injury in sports-related concussion (SRC) requires understanding the neural mechanisms involved during the time of recovery after injury. The decision for return-to-play is one of the most difficult responsibilities facing the physician, and so far this decision has been based primarily on neurological examination, symptom checklists, and neuropsychological (NP) testing. Functional magnetic resonance imaging (fMRI) may be an additional, more objective tool to assess the severity and recovery of function after concussion. The purpose of this study was to define neural correlates of SRC during the 2 months after injury in varsity contact sport athletes who suffered a SRC. All athletes were scanned as they performed an n-back task, for n=1, 2, 3. Subjects were scanned within 72 hours (session one), at 2 weeks (session two), and 2 months (session three) post-injury. Compared with age and sex matched normal controls, concussed subjects demonstrated persistent, significantly increased activation for the 2 minus 1 n-back contrast in bilateral dorsolateral prefrontal cortex (DLPFC) in all three sessions and in the inferior parietal lobe in session one and two (α≤0.01 corrected). Measures of task performance revealed no significant differences between concussed versus control groups at any of the three time points with respect to any of the three n-back tasks. These findings suggest that functional brain activation differences persist at 2 months after injury in concussed athletes, despite the fact that their performance on a standard working memory task is comparable to normal controls and normalization of clinical and NP test results. These results might indicate a delay between neural and behaviorally assessed recovery after SRC.

Automatic cropping of MRI rat brain volumes using pulse coupled neural networks.

The Pulse Coupled Neural Network (PCNN) was developed by Eckhorn to model the observed synchronization of neural assemblies in the visual cortex of small mammals such as a cat. In this paper we show the use of the PCNN as an image segmentation strategy to crop MR images of rat brain volumes. We then show the use of the associated PCNN image 'signature' to automate the brain cropping process with a trained artificial neural network. We tested this novel algorithm on three T2 weighted acquisition configurations comprising a total of 42 rat brain volumes. The datasets included 40 ms, 48 ms and 53 ms effective TEs, acquisition field strengths of 4.7 T and 9.4 T, image resolutions from 64x64 to 256x256, slice locations ranging from +6 mm to -11 mm AP, two different surface coil manufacturers and imaging protocols. The results were compared against manually segmented gold standards and Brain Extraction Tool (BET) V2.1 results. The Jaccard similarity index was used for numerical evaluation of the proposed algorithm. Our novel PCNN cropping system averaged 0.93 compared to BET scores circa 0.84.

Imaging the neural circuitry and chemical control of aggressive motivation.

BACKGROUND: With the advent of functional magnetic resonance imaging (fMRI) in awake animals it is possible to resolve patterns of neuronal activity across the entire brain with high spatial and temporal resolution. Synchronized changes in neuronal activity across multiple brain areas can be viewed as functional neuroanatomical circuits coordinating the thoughts, memories and emotions for particular behaviors. To this end, fMRI in conscious rats combined with 3D computational analysis was used to identifying the putative distributed neural circuit involved in aggressive motivation and how this circuit is affected by drugs that block aggressive behavior. RESULTS: To trigger aggressive motivation, male rats were presented with their female cage mate plus a novel male intruder in the bore of the magnet during image acquisition. As expected, brain areas previously identified as critical in the organization and expression of aggressive behavior were activated, e.g., lateral hypothalamus, medial basal amygdala. Unexpected was the intense activation of the forebrain cortex and anterior thalamic nuclei. Oral administration of a selective vasopressin V1a receptor antagonist SRX251 or the selective serotonin reuptake inhibitor fluoxetine, drugs that block aggressive behavior, both caused a general suppression of the distributed neural circuit involved in aggressive motivation. However, the effect of SRX251, but not fluoxetine, was specific to aggression as brain activation in response to a novel sexually receptive female was unaffected. CONCLUSION: The putative neural circuit of aggressive motivation identified with fMRI includes neural substrates contributing to emotional expression (i.e. cortical and medial amygdala, BNST, lateral hypothalamus), emotional experience (i.e. hippocampus, forebrain cortex, anterior cingulate, retrosplenial cortex) and the anterior thalamic nuclei that bridge the motor and cognitive components of aggressive responding. Drugs that block vasopressin neurotransmission or enhance serotonin activity suppress activity in this putative neural circuit of aggressive motivation, particularly the anterior thalamic nuclei.