Automated cardiac coverage assessment in cardiovascular magnetic resonance imaging using an explainable recurrent 3D dual-domain convolutional network.

BACKGROUND: Cardiovascular magnetic resonance (CMR) imaging has become a modality with superior power for the diagnosis and prognosis of cardiovascular diseases. One of the essential quality controls of CMR images is to investigate the complete cardiac coverage, which is necessary for the volumetric and functional assessment. PURPOSE: This study examines the full cardiac coverage using a 3D dual-domain convolutional model and then improves this model using an innovative explainable salient region detection model and a recurrent architecture. METHODS: Salient regions are extracted from the short-axis cine CMR stacks using a three-step proposed algorithm. Changing the architecture of the 3D dual-domain convolutional model to a recurrent one and taking advantage of the salient region detection model creates a kind of attention mechanism that leads to improved results. RESULTS: The results obtained from the images of over 6200 participants of the UK Biobank population cohort study show the superiority of the proposed model over the previous studies. The dataset is the largest regarding the number of participants to control the cardiac coverage. The accuracies of the proposed model in identifying the presence/absence of basal/apical slices are 96.22% and 95.42%, respectively. CONCLUSION: The proposed recurrent architecture of the 3D dual-domain convolutional model can force the model to focus on the most informative areas of the images using the extracted salient regions, which can help the model improve accuracy. The performance of the proposed fully automated model indicates that it can be used for image quality control in population cohort datasets and real-time post-imaging quality assessments. Codes are available at https://github.com/mohammadhashemii/CMR_Cardiac_Coverage_Control.

Deep reinforcement learning in continuous action space for autonomous robotic surgery.

PURPOSE: Reinforcement learning methods have shown promising results for the automation of sub-tasks in robotic surgery systems. With the development of these methods, surgical robots have been able to achieve good performances, so that they can be used in complex and high-risk environments such as surgical pattern cutting to reduce stress and pressure on the surgeon and increase surgical accuracy. This study has aimed at providing a deep reinforcement learning-based approach to control the gripper arm when cutting soft tissue in a continuous action space. METHODS: Surgical soft tissue cutting in this study is performed by controlling the gripper arm in a continuous action space and a grid observation space. In the proposed method using deep reinforcement learning, we find an optimal tensioning policy in the continuous action space that increases the cutting accuracy of the predetermined pattern. RESULTS: The simulation results demonstrated that in the cutting of many complex patterns, the proposed method works better than the methods in which the tensioning was performed in a discrete action space and the observation space was modeled as a partial and random representation. CONCLUSION: We introduced a deep reinforcement learning-based method for obtaining the optimal tensioning policy in a continuous action space when cutting a predetermined pattern. We showed that the proposed approach outperforms the state-of-the-art method in the soft pattern cutting task with respect to accuracy.

A generalised deep meta-learning model for automated quality control of cardiovascular magnetic resonance images.

BACKGROUND AND OBJECTIVES: Cardiovascular magnetic resonance (CMR) imaging is a powerful modality in functional and anatomical assessment for various cardiovascular diseases. Sufficient image quality is essential to achieve proper diagnosis and treatment. A large number of medical images, the variety of imaging artefacts, and the workload of imaging centres are amongst the factors that reveal the necessity of automatic image quality assessment (IQA). However, automated IQA requires access to bulk annotated datasets for training deep learning (DL) models. Labelling medical images is a tedious, costly and time-consuming process, which creates a fundamental challenge in proposing DL-based methods for medical applications. This study aims to present a new method for CMR IQA when there is limited access to annotated datasets. METHODS: The proposed generalised deep meta-learning model can evaluate the quality by learning tasks in the prior stage and then fine-tuning the resulting model on a small labelled dataset of the desired tasks. This model was evaluated on the data of over 6,000 subjects from the UK Biobank for five defined tasks, including detecting respiratory motion, cardiac motion, Aliasing and Gibbs ringing artefacts and images without artefacts. RESULTS: The results of extensive experiments show the superiority of the proposed model. Besides, comparing the model's accuracy with the domain adaptation model indicates a significant difference by using only 64 annotated images related to the desired tasks. CONCLUSION: The proposed model can identify unknown artefacts in images with acceptable accuracy, which makes it suitable for medical applications and quality assessment of large cohorts. CODE AVAILABILITY: https://github.com/HosseinSimchi/META-IQA-CMRImages.

Prediction of Acute Kidney Injury After Cardiac Surgery Using Interpretable Machine Learning.

Background: Acute kidney injury (AKI) is a complication that occurs for various reasons after surgery, especially cardiac surgery. This complication can lead to a prolonged treatment process, increased costs, and sometimes death. Prediction of postoperative AKI can help anesthesiologists to implement preventive and early treatment strategies to reduce the risk of AKI. Objectives: This study tries to predict postoperative AKI using interpretable machine learning models. Methods: For this study, the information of 1435 patients was collected from multiple centers. The gathered data are in six categories: demographic characteristics and type of surgery, past medical history (PMH), drug history (DH), laboratory information, anesthesia and surgery information, and postoperative variables. Machine learning methods, including support vector machine (SVM), multilayer perceptron (MLP), decision tree (DT), random forest (RF), logistic regression, XGBoost, and AdaBoost, were used to predict postoperative AKI. Local interpretable model-agnostic explanations (LIME) and the Shapley methods were then leveraged to check the interpretability of models. Results: Comparing the area under the curves (AUCs) obtained for different machine learning models show that the RF and XGBoost methods with values of 0.81 and 0.80 best predict postoperative AKI. The interpretations obtained for the machine learning models show that creatinine (Cr), cardiopulmonary bypass time (CPB time), blood sugar (BS), and albumin (Alb) have the most significant impact on predictions. Conclusions: The treatment team can be informed about the possibility of postoperative AKI before cardiac surgery using machine learning models such as RF and XGBoost and adjust the treatment procedure accordingly. Interpretability of predictions for each patient ensures the validity of obtained predictions.

Medical imaging and computational image analysis in COVID-19 diagnosis: A review.

Coronavirus disease (COVID-19) is an infectious disease caused by a newly discovered coronavirus. The disease presents with symptoms such as shortness of breath, fever, dry cough, and chronic fatigue, amongst others. The disease may be asymptomatic in some patients in the early stages, which can lead to increased transmission of the disease to others. This study attempts to review papers on the role of imaging and medical image computing in COVID-19 diagnosis. For this purpose, PubMed, Scopus and Google Scholar were searched to find related studies until the middle of 2021. The contribution of this study is four-fold: 1) to use as a tutorial of the field for both clinicians and technologists, 2) to comprehensively review the characteristics of COVID-19 as presented in medical images, 3) to examine automated artificial intelligence-based approaches for COVID-19 diagnosis, 4) to express the research limitations in this field and the methods used to overcome them. Using machine learning-based methods can diagnose the disease with high accuracy from medical images and reduce time, cost and error of diagnostic procedure. It is recommended to collect bulk imaging data from patients in the shortest possible time to improve the performance of COVID-19 automated diagnostic methods.

Deep learning approaches for automated classification and segmentation of head and neck cancers and brain tumors in magnetic resonance images: a meta-analysis study.

PURPOSE: Deep learning (DL) has led to widespread changes in automated segmentation and classification for medical purposes. This study is an attempt to use statistical methods to analyze studies related to segmentation and classification of head and neck cancers (HNCs) and brain tumors in MRI images. METHODS: PubMed, Web of Science, Embase, and Scopus were searched to retrieve related studies published from January 2016 to January 2020. Studies that evaluated the performance of DL-based models in the segmentation, and/or classification and/or grading of HNCs and/or brain tumors were included. Selected studies for each analysis were statistically evaluated based on the diagnostic performance metrics. RESULTS: The search results retrieved 1,664 related studies, of which 30 studies were eligible for meta-analysis. The overall performance of DL models for the complete tumor in terms of the pooled Dice score, sensitivity, and specificity was 0.8965 (95% confidence interval (95% CI): 0.76-0.9994), 0.9132 (95% CI: 0.71-0.994) and 0.9164 (95% CI: 0.78-1.00), respectively. The DL methods achieved the highest performance for classifying three types of glioma, meningioma, and pituitary tumors with overall accuracies of 96.01%, 99.73%, and 96.58%, respectively. Stratification of glioma tumors by high and low grading revealed overall accuracies of 94.32% and 94.23% for the DL methods, respectively. CONCLUSION: Based on the obtained results, we can acknowledge the significant ability of DL methods in the mentioned applications. Poor reporting in these studies challenges the analysis process, so it is recommended that future studies report comprehensive results based on different metrics.

A Random Walk Approach to Query Informative Constraints for Clustering.

This paper presents a random walk approach to the problem of querying informative constraints for clustering. The proposed method is based on the properties of the commute time, that is the expected time taken for a random walk to travel between two nodes and return, on the adjacency graph of data. Commute time has the nice property of that, the more short paths connect two given nodes in a graph, the more similar those nodes are. Since computing the commute time takes the Laplacian eigenspectrum into account, we use this property in a recursive fashion to query informative constraints for clustering. At each recursion, the proposed method constructs the adjacency graph of data and utilizes the spectral properties of the commute time matrix to bipartition the adjacency graph. Thereafter, the proposed method benefits from the commute times distance on graph to query informative constraints between partitions. This process iterates for each partition until the stop condition becomes true. Experiments on real-world data show the efficiency of the proposed method for constraints selection.

Querying Beneficial Constraints Before Clustering Using Facility Location Analysis.

This paper examines the problem of querying beneficial constraints before clustering. Existing methods in this area choose constraints heuristically based on some prior assumptions on the usefulness of constraints. However, the usefulness and propagation of constraints are two important issues in the constraints selection that are not investigated simultaneously in most existing works. This paper addresses the problem of querying beneficial constraints using facility location analysis that is one of the most well-studied areas of the operations research. To this end, the source problem of querying beneficial constraints is transformed into an instance of target uncapacitated -facility location problem ( -UFL) and then is benefited from existing algorithms in the target space to find a solution to the -UFL problem. The solution to the -UFL problem is then transformed into a solution of the source problem of querying beneficial constraints. Both usefulness and propagation of constraints are achieved in this paper by respectively mapping them into the corresponding opening and service costs in target problem space and then minimizing the total cost in target space. The proposed method is based on an optimization framework and is entirely different from existing methods in the constraints selection that are limited to greedy approaches. A range of experiments is presented to compare the proposed method to alternatives and explore its behavior in the selection of clustering constraints.