DARQ: Deep learning of quality control for stereotaxic registration of human brain MRI

Linear registration to stereotaxic space is a common first step in many automated image-processing tools for analysis of human brain MRI scans. This step is crucial for the success of the subsequent image-processing steps. Several well-established algorithms are commonly used in the field of neuroimaging for this task, but none have a 100% success rate. Manual assessment of the registration is commonly used as part of quality control. To reduce the burden of this time-consuming step, we propose Deep Automated Registration Qc (DARQ), a fully automatic quality control method based on deep learning that can replace the human rater and accurately perform quality control assessment for stereotaxic registration of T1w brain scans. In a recently published study from our group comparing linear registration methods, we used a database of 9325 MRI scans from several publicly available datasets and applied seven linear registration tools to them. In this study, the resulting images that were assessed and labeled by a human rater are used to train a deep neural network to detect cases when registration failed. We further validated the results on an independent dataset of patients with multiple sclerosis, with manual QC labels available (n=1200). In terms of agreement with a manual rater, our automated QC method was able to achieve 89% accuracy and 85% true negative rate (equivalently 15% false positive rate) in detecting scans that should pass quality control in a balanced cross-validation experiments, and 96.1% accuracy and 95.5% true negative rate (or 4.5% FPR) when evaluated in a balanced independent sample, similar to manual QC rater (test-retest accuracy of 93%). The results show that DARQ is robust, fast, accurate, and generalizable in detecting failure in linear stereotaxic registrations and can substantially reduce QC time (by a factor of 20 or more) when processing large datasets.

Logistic versus linear regression-based Reliable Change Index: implications for clinical studies with diverse sample sizes

The linear regression-based Reliable Change Index (RCI) is widely used to identify memory impairments through longitudinal assessment. However, the minimum sample size required for estimates to be reliable has never been specified. Using the Alzheimer’s Disease Neuroimaging Initiative data as true parameters, we run simulations for samples of size 10 to 1000 and analyzed the percentage of times the estimates are significant, their coverage rate, and the accuracy of the models including both the True Positive Rate (TPR) and the True Negative Rate (TNR). We compared the linear RCI with a logistic RCI for discrete, bounded scores. We found that the logistic RCI is more accurate than the linear RCI overall, with the linear RCI approximating the logistic RCI for samples of size 200 or greater. We provide an R code for researchers and clinicians to calculate the logistic RCI with samples smaller than 200.

One Shot Model For The Prediction of COVID-19 and Lesions Segmentation In Chest CT Scans Through The Affinity Among Lesion Mask Features

We introduce a model that segments lesions and predicts COVID-19 from chest CT scans through the derivation of an affinity matrix between lesion masks. The novelty of the methodology is based on the computation of the affinity between the lesion masks’ features extracted from the image. First, a batch of vectorized lesion masks is constructed. Then, the model learns the parameters of the affinity matrix that captures the relationship between features in each vector. Finally, the affinity is expressed as a single vector of pre-defined length. Without any complicated data manipulation, class balancing tricks, and using only a fraction of the training data, we achieve a 91.74% COVID-19 sensitivity, 85.35% common pneumonia sensitivity, 97.26% true negative rate and 91.94% F1-score. Ablation studies show that the method can quickly generalize to new datasets. All source code, models and results are publicly available on https://github.com/AlexTS1980/COVID-Affinity-Model.

Detection and Segmentation of Lesion Areas in Chest CT Scans For The Prediction of COVID-19

In this paper we compare the models for the detection and segmentation of Ground Glass Opacity and Consolidation in chest CT scans. These lesion areas are often associated both with common pneumonia and COVID-19. We train a Mask R-CNN model to segment these areas with high accuracy using three approaches: merging masks for these lesions into one, deleting the mask for Consolidation, and using both masks separately. The best model achieves the mean average precision of 44.68% using MS COCO criterion for instance segmentation across all accuracy thresholds. The classification model, COVID-CT-Mask-Net, which learns to predict the presence of COVID-19 vs common pneumonia vs control, achieves the 93.88% COVID-19 sensitivity, 95.64% overall accuracy, 95.06% common pneumonia sensitivity and 96.91% true negative rate on the COVIDx-CT test split (21192 CT scans) using a small fraction of the training data. We also analyze the effect of Non-Maximum Suppression of overlapping object predictions, both on the segmentation and classification accuracy. The full source code, models and pretrained weights are available on https://github.com/AlexTS1980/COVID-CT-Mask-Net.

Fast and Robust Algorithm for Detecting Body Posture Using Wrist-Worn Accelerometers

Background: The increasing popularity of wrist-worn accelerometers introduces novel challenges to the research on physical activity and sedentary behavior. Estimation of body posture is one such challenge. Methods: The authors proposed an approach called SedUp to differentiate between sedentary (sitting/lying) and standing postures. SedUp is based on the logistic regression classifier, using the wrist elevation and the motion variability extracted from raw accelerometry data collected on the axis parallel to the forearm. The authors developed and tested our method on data from N = 45 community-dwelling older adults. All subjects wore ActiGraph GT3X+ accelerometers on the left and right wrist, and activPAL was placed on the thigh in the free-living environment for 7 days. ActivPAL provided ground truth about body posture. The authors reported SedUp’s classification accuracy for each wrist separately. Results: Using the data from the left wrist, SedUp estimated the standing posture with median true positive rate = 0.83 and median true negative rate = 0.91. Using the data from the right wrist, SedUp estimated the standing posture with median true positive rate = 0.86 and median true negative rate = 0.93. Conclusions: SedUp provides accurate classification of body posture using wrist-worn accelerometers. The separate validation for each wrist allows for the application of SedUp in a wide spectrum of free-living studies.

Detection and Segmentation of Lesion Areas in Chest CT Scans For The Prediction of COVID-19

In this paper we compare the models for the detection and segmentation of Ground Glass Opacity and Consolidation in chest CT scans. These lesion areas are often associated both with common pneumonia and COVID-19. We train a Mask R-CNN model to segment these areas with high accuracy using three approaches: merging masks for these lesions into one, deleting the mask for Consolidation, and using both masks separately. The best model achieves the mean average precision of 44.68% using MS COCO criterion for instance segmentation across all accuracy thresholds. The classification model, COVID-CT-Mask-Net, which learns to predict the presence of COVID-19 vs common pneumonia vs control, achieves the 93.88% COVID-19 sensitivity, 95.64% overall accuracy, 95.06% common pneumonia sensitivity and 96.91% true negative rate on the COVIDx-CT test split (21192 CT scans) using a small fraction of the training data. We also analyze the effect of Non-Maximum Suppression of overlapping object predictions, both on the segmentation and classification accuracy. The full source code, models and pretrained weights are available on https://github.com/AlexTS1980/COVID-CT-Mask-Net.

Detection and Segmentation of Lesion Areas in Chest CT Scans For The Prediction of COVID-19

In this paper we compare the models for the detection and segmentation of Ground Glass Opacity and Consolidation in chest CT scans. These lesion areas are often associated both with common pneumonia and COVID-19. We train a Mask R-CNN model to segment these areas with high accuracy using three approaches: merging masks for these lesions into one, deleting the mask for Consolidation, and using both masks separately. The best model achieves the mean average precision of 44.68% using MS COCO criterion for instance segmentation across all accuracy thresholds. The classification model, COVID-CT-Mask-Net, which learns to predict the presence of COVID-19 vs common pneumonia vs control, achieves the 93.88% COVID-19 sensitivity, 95.64% overall accuracy, 95.06% common pneumonia sensitivity and 96.91% true negative rate on the COVIDx-CT test split (21192 CT scans) using a small fraction of the training data. We also analyze the effect of Non-Maximum Suppression of overlapping object predictions, both on the segmentation and classification accuracy. The full source code, models and pretrained weights are available on https://github.com/AlexTS1980/COVID-CT-Mask-Net.

An Attention-Based Network for Textured Surface Anomaly Detection

Textured surface anomaly detection is a significant task in industrial scenarios. In order to further improve the detection performance, we proposed a novel two-stage approach with an attention mechanism. Firstly, in the segmentation network, the feature extraction and anomaly attention modules are designed to capture the detail information as much as possible and focus on the anomalies, respectively. To strike dynamic balances between these two parts, an adaptive scheme where learnable parameters are gradually optimized is introduced. Subsequently, the weights of the segmentation network are frozen, and the outputs are fed into the classification network, which is trained independently in this stage. Finally, we evaluate the proposed approach on DAGM 2007 dataset which consists of diverse textured surfaces with weakly-labeled anomalies, and the experiments demonstrate that our method can achieve 100% detection rates in terms of TPR (True Positive Rate) and TNR (True Negative Rate).

A multi-parameter diagnostic clinical decision tree for the rapid diagnosis of tuberculosis in HIV-positive patients presenting to an emergency centre

Background: Early diagnosis is essential to reduce the morbidity and mortality of HIV-associated tuberculosis. We developed a multi-parameter clinical decision tree to facilitate rapid diagnosis of tuberculosis using point-of-care diagnostic tests in HIV-positive patients presenting to an emergency centre. Methods: A cross-sectional study was performed in a district hospital emergency centre in a high-HIV-prevalence community in South Africa. Consecutive HIV-positive adults with ≥1 WHO tuberculosis symptoms were enrolled over a 16-month period. Point-of-care ultrasound (PoCUS) and urine lateral flow lipoarabinomannan (LF-LAM) assay were done according to standardized protocols. Participants also received a chest X-ray. Reference standard was the detection of Mycobacterium tuberculosis using Xpert MTB/RIF or culture. Logistic regressions models were used to investigate the independent association between prevalent microbiologically confirmed tuberculosis and clinical and biological variables of interest. A decision tree model to predict tuberculosis was developed using the classification and regression tree algorithm. Results: There were 414 participants enrolled: 171 male, median age 36 years, median CD4 cell count 86 cells/mm3. Tuberculosis prevalence was 42% (n=172). Significant variables used to build the classification tree included ≥2 WHO symptoms, antiretroviral therapy use, LF-LAM, PoCUS independent features (pericardial effusion, ascites, intra-abdominal lymphadenopathy) and chest X-ray. LF-LAM was positioned after WHO symptoms (75% true positive rate, representing 17% of study population). Chest X-ray should be performed next if LF-LAM is negative. The presence of ≤1 PoCUS independent feature in those with ‘possible or unlikely tuberculosis’ on chest x-ray represented 47% of non-tuberculosis participants (true negative rate 83%). In a prediction tree which only included true point-of-care tests, a negative LF-LAM and the presence of ≤2 independent PoCUS features had a 71% true negative rate (representing 53% of sample). Conclusions: LF-LAM should be performed in all adults with suspected HIV-associated tuberculosis (regardless of CD4 cell count) presenting to the emergency centre.