Top-performing Approaches Research Articles

Generating Contrast-Enhanced Liver MRI Images from Native Sequences

Abstract Generating synthetic contrast-enhanced liver MRI scans from native MRI images can serve to mitigate the issue of sparse contrast-enhanced image datasets while concurrently circumventing the time-consuming and costly process of administering contrast agents during image acquisition. In this study, we conducted three experiments using paired image-to-image translation techniques. Native T1 sequences showing the abdominal liver region served as the input, while contrast-enhanced T1 sequences were the target. The data preprocessing methods and image boundaries were varied for the individual experiments in addition to the implementation of a 5-fold cross-validation for the top-performing approach. Focusing on liver regions achieved the best results with a mean absolute error (MAE) of 0.0837 ± 0.0068, a mean squared error (MSE) of 0.0128 ± 0.0023 and a peak signal-to-noise ratio (PSNR) of 18.99 ± 0.81 dB. Our findings serve as a proof-of-concept, demonstrating the feasibility of generating contrast-enhanced MRI images. However, the current state necessitates further enhancements for effectively addressing the challenge posed by limited dataset sizes with difficult anatomical circumstances as well as MR imaging-related heterogenous tissue contrasts.

Spatial modelling and classification of altered coal using random forest-based methods at Moatize Basin, Mozambique

Coal basins affected by intrusion pose distinctive challenges, such as the complexity of determining the extent of impacted coal from borehole cuttings and the potential for inaccuracies in the spatial modelling of coal quality parameters. It is more challenging in complex thinly bedded coal seams, characterised by multiple sub-layers of varying thickness and coal quality. This study employs machine learning techniques in the highly complex thinly bedded Permian-aged Chipanga coal seam in the Moatize Basin, Mozambique. The study aims to assess the efficacy of these techniques for the spatial modelling and classification of altered coals. The results demonstrate that the random forest-based classifier can discriminate between altered and fresh coal with a range of accuracy of 97%–100%. Furthermore, among the five spatial methods assessed, geographical quantile regression forest emerges as one of the top-performing approaches for assessing the impact of magmatic intrusions on coal properties within the Moatize Basin. Although previous studies have shown the ability of these methods for spatial modelling and coal classification on single coal seam, these techniques have not been used in complex thin-bedded coal seams. Therefore, the findings of this study highlight the implications of these methods in navigating the complexities of thin-bedded coal seams.

Automated real-time prediction of geological formation tops during drilling operations: an applied machine learning solution for the Norwegian Continental Shelf

Accurate prediction of geological formation tops is a crucial task for optimizing hydrocarbon exploration and production activities. This research investigates and conducts a comprehensive comparative analysis of several advanced machine learning approaches tailored for the critical application of geological formation top prediction within the complex Norwegian Continental Shelf (NCS) region. The study evaluates and benchmarks the performance of four prominent machine learning models: Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest ensemble method, and Multi-Layer Perceptron (MLP) neural network. To facilitate a rigorous assessment, the models are extensively evaluated across two distinct datasets - a dedicated test dataset and a blind dataset independent for validation. The evaluation criteria revolve around quantifying the models' predictive accuracy in successfully classifying multiple geological formation top types. Additionally, the study employs the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm as a baseline benchmarking technique to contextualize the relative performance of the machine learning models against a conventional clustering approach. Leveraging two model-agnostic feature importance analysis techniques - Permutation Feature Importance (PFI) and Shapley Additive exPlanations (SHAP), the investigation identifies and ranks the most influential input variables driving the predictive capabilities of the models. The comprehensive analysis unveils the MLP neural network model as the top-performing approach, achieving remarkable predictive accuracy with a perfect score of 0.99 on the blind validation dataset, surpassing the other machine learning techniques as well as the DBSCAN benchmark. However, the SVM model attains superior performance on the initial test dataset, with an accuracy of 0.99. Intriguingly, the PFI and SHAP analyses converge in consistently pinpointing depth (DEPT), revolution per minute (RPM), and Hook-load (HKLD) as the three most impactful parameters influencing model predictions across the different algorithms. These findings underscore the potential of sophisticated machine learning methodologies, particularly neural network-based models, to significantly enhance the accuracy of geological formation top prediction within the geologically complex NCS region. However, the study emphasizes the necessity for further extensive testing on larger datasets to validate the generalizability of the high performance observed. Overall, this research delivers an exhaustive comparative evaluation of state-of-the-art machine learning techniques, offering critical insights to guide the optimal selection, development, and real-world deployment of accurate and reliable predictive modeling strategies tailored for hydrocarbon exploration and reservoir characterization endeavors in the NCS.Graphical abstract

Reproducibility of arterial spin labeling cerebral blood flow image processing: A report of the ISMRM open science initiative for perfusion imaging (OSIPI) and the ASL MRI challenge.

Arterial spin labeling (ASL) is a widely used contrast-free MRI method for assessing cerebral blood flow (CBF). Despite the generally adopted ASL acquisition guidelines, there is still wide variability in ASL analysis. We explored this variability through the ISMRM-OSIPI ASL-MRI Challenge, aiming to establish best practices for more reproducible ASL analysis. Eight teams analyzed the challenge data, which included a high-resolution T1-weighted anatomical image and 10 pseudo-continuous ASL datasets simulated using a digital reference object to generate ground-truth CBF values in normal and pathological states. We compared the accuracy of CBF quantification from each team's analysis to the ground truth across all voxels and within predefined brain regions. Reproducibility of CBF across analysis pipelines was assessed using the intra-class correlation coefficient (ICC), limits of agreement (LOA), and replicability of generating similar CBF estimates from different processing approaches. Absolute errors in CBF estimates compared to ground-truth synthetic data ranged from 18.36 to 48.12 mL/100 g/min. Realistic motion incorporated into three datasets produced the largest absolute error and variability between teams, with the least agreement (ICC and LOA) with ground-truth results. Fifty percent of the submissions were replicated, and one produced three times larger CBF errors (46.59 mL/100 g/min) compared to submitted results. Variability in CBF measurements, influenced by differences in image processing, especially to compensate for motion, highlights the significance of standardizing ASL analysis workflows. We provide a recommendation for ASL processing based on top-performing approaches as a step toward ASL standardization.

Automatic piecewise linear regression

Regression modelling often presents a trade-off between predictiveness and interpretability. Highly predictive and popular tree-based algorithms such as Random Forest and boosted trees predict very well the outcome of new observations, but the effect of the predictors on the result is hard to interpret. Highly interpretable algorithms like linear effect-based boosting and MARS, on the other hand, are typically less predictive. Here we propose a novel regression algorithm, automatic piecewise linear regression (APLR), that combines the predictiveness of a boosting algorithm with the interpretability of a MARS model. In addition, as a boosting algorithm, it automatically handles variable selection, and, as a MARS-based approach, it takes into account non-linear relationships and possible interaction terms. We show on simulated and real data examples how APLR’s performance is comparable to that of the top-performing approaches in terms of prediction, while offering an easy way to interpret the results. APLR has been implemented in C++ and wrapped in a Python package as a Scikit-learn compatible estimator.

Solving the single depot open close multiple travelling salesman problem through a multi-chromosome based genetic algorithm

The multiple travelling salesman problem (MTSP) extends the classical travelling salesman problem (TSP) by involving multiple salesman in the solution. MTSP has found widespread applications in various domains, such as transportation, robotics, and networking. Despite extensive research on MTSP and its variants, there has been limited attention given to the open close multiple travelling salesman problem (OCMTSP) and its variants in the literature. To the best of the author's knowledge, only one study has addressed OCMTSP, introducing an exact algorithm designed for optimal solutions. However, the efficiency of this existing algorithm diminishes for larger instances due to computational complexity. Therefore, there is a crucial need for a high-level metaheuristic to provide optimal/best solutions within a reasonable timeframe. Addressing this gap, this study proposes a first meta-heuristic called multi-chromosome-based Genetic Algorithm (GA) for solving OCMTSP. The effectiveness of the developed algorithm is demonstrated through a comparative study on distinct asymmetric benchmark instances sourced from the TSPLIB dataset. Additionally, results from comprehensive experiments conducted on 90 OCMTSP symmetric instances, generated from the renowned TSPLIB benchmark dataset, highlight the efficiency of the proposed GA in addressing the OCMTSP. Notably, the proposed multi-chromosome-based GA stands out as the top-performing approach in terms of overall performance. Further, solutions to symmetric TSPLIB benchmark instances are also reported, which will be used as a basis for future studies.

Current Progress and Challenges in Large-Scale 3D Mitochondria Instance Segmentation.

In this paper, we present the results of the MitoEM challenge on mitochondria 3D instance segmentation from electron microscopy images, organized in conjunction with the IEEE-ISBI 2021 conference. Our benchmark dataset consists of two large-scale 3D volumes, one from human and one from rat cortex tissue, which are 1,986 times larger than previously used datasets. At the time of paper submission, 257 participants had registered for the challenge, 14 teams had submitted their results, and six teams participated in the challenge workshop. Here, we present eight top-performing approaches from the challenge participants, along with our own baseline strategies. Posterior to the challenge, annotation errors in the ground truth were corrected without altering the final ranking. Additionally, we present a retrospective evaluation of the scoring system which revealed that: 1) challenge metric was permissive with the false positive predictions; and 2) size-based grouping of instances did not correctly categorize mitochondria of interest. Thus, we propose a new scoring system that better reflects the correctness of the segmentation results. Although several of the top methods are compared favorably to our own baselines, substantial errors remain unsolved for mitochondria with challenging morphologies. Thus, the challenge remains open for submission and automatic evaluation, with all volumes available for download.

Identification of disease modules using higher-order network structure.

Higher-order interaction patterns among proteins have the potential to reveal mechanisms behind molecular processes and diseases. While clustering methods are used to identify functional groups within molecular interaction networks, these methods largely focus on edge density and do not explicitly take into consideration higher-order interactions. Disease genes in these networks have been shown to exhibit rich higher-order structure in their vicinity, and considering these higher-order interaction patterns in network clustering have the potential to reveal new disease-associated modules. We propose a higher-order community detection method which identifies community structure in networks with respect to specific higher-order connectivity patterns beyond edges. Higher-order community detection on four different protein-protein interaction networks identifies biologically significant modules and disease modules that conventional edge-based clustering methods fail to discover. Higher-order clusters also identify disease modules from genome-wide association study data, including new modules that were not discovered by top-performing approaches in a Disease Module DREAM Challenge. Our approach provides a more comprehensive view of community structure that enables us to predict new disease-gene associations. https://github.com/Reed-CompBio/graphlet-clustering.

Evaluating the utility of identity-by-descent segment numbers for relatedness inference via information theory and classification.

Despite decades of methods development for classifying relatives in genetic studies, pairwise relatedness methods’ recalls are above 90% only for first through third-degree relatives. The top-performing approaches, which leverage identity-by-descent segments, often use only kinship coefficients, while others, including estimation of recent shared ancestry (ERSA), use the number of segments relatives share. To quantify the potential for using segment numbers in relatedness inference, we leveraged information theory measures to analyze exact (i.e. produced by a simulator) identity-by-descent segments from simulated relatives. Over a range of settings, we found that the mutual information between the relatives’ degree of relatedness and a tuple of their kinship coefficient and segment number is on average 4.6% larger than between the degree and the kinship coefficient alone. We further evaluated identity-by-descent segment number utility by building a Bayes classifier to predict first through sixth-degree relationships using different feature sets. When trained and tested with exact segments, the inclusion of segment numbers improves the recall by between 0.28% and 3% for second through sixth-degree relatives. However, the recalls improve by less than 1.8% per degree when using inferred segments, suggesting limitations due to identity-by-descent detection accuracy. Last, we compared our Bayes classifier that includes segment numbers with both ERSA and IBIS and found comparable recalls, with the Bayes classifier and ERSA slightly outperforming each other across different degrees. Overall, this study shows that identity-by-descent segment numbers can improve relatedness inference, but errors from current SNP array-based detection methods yield dampened signals in practice.

Estimating the optimal individualized treatment rule from a cost-effectiveness perspective.

Optimal individualized treatment rules (ITRs) provide customized treatment recommendations based on subject characteristics to maximize clinical benefit in accordance with the objectives in precision medicine. As a result, there is growing interest in developing statistical tools for estimating optimal ITRs in evidence-based research. In health economic perspectives, policy makers consider the tradeoff between health gains and incremental costs of interventions to set priorities and allocate resources. However, most work on ITRs has focused on maximizing the effectiveness of treatment without considering costs. In this paper, we jointly consider the impact of effectiveness and cost on treatment decisions and define ITRs under a composite-outcome setting, so that we identify the most cost-effective ITR that accounts for individual-level heterogeneity through direct optimization. In particular, we propose a decision-tree-based statistical learning algorithm that uses a net-monetary-benefit-based reward to provide nonparametric estimations of the optimal ITR. We provide several approaches to estimating the reward underlying the ITR as a function of subject characteristics. We present the strengths and weaknesses of each approach and provide practical guidelines by comparing their performance in simulation studies. We illustrate the top-performing approach from our simulations by evaluating the projected 15-year personalized cost-effectiveness of the intensive blood pressure control of the Systolic Blood Pressure Intervention Trial (SPRINT)study.

AGE challenge: Angle Closure Glaucoma Evaluation in Anterior Segment Optical Coherence Tomography.

Angle closure glaucoma (ACG) is a more aggressive disease than open-angle glaucoma, where the abnormal anatomical structures of the anterior chamber angle (ACA) may cause an elevated intraocular pressure and gradually lead to glaucomatous optic neuropathy and eventually to visual impairment and blindness. Anterior Segment Optical Coherence Tomography (AS-OCT) imaging provides a fast and contactless way to discriminate angle closure from open angle. Although many medical image analysis algorithms have been developed for glaucoma diagnosis, only a few studies have focused on AS-OCT imaging. In particular, there is no public AS-OCT dataset available for evaluating the existing methods in a uniform way, which limits progress in the development of automated techniques for angle closure detection and assessment. To address this, we organized the Angle closure Glaucoma Evaluation challenge (AGE), held in conjunction with MICCAI 2019. The AGE challenge consisted of two tasks: scleral spur localization and angle closure classification. For this challenge, we released a large dataset of 4800 annotated AS-OCT images from 199 patients, and also proposed an evaluation framework to benchmark and compare different models. During the AGE challenge, over 200 teams registered online, and more than 1100 results were submitted for online evaluation. Finally, eight teams participated in the onsite challenge. In this paper, we summarize these eight onsite challenge methods and analyze their corresponding results for the two tasks. We further discuss limitations and future directions. In the AGE challenge, the top-performing approach had an average Euclidean Distance of 10 pixels (10µm) in scleral spur localization, while in the task of angle closure classification, all the algorithms achieved satisfactory performances, with two best obtaining an accuracy rate of 100%. These artificial intelligence techniques have the potential to promote new developments in AS-OCT image analysis and image-based angle closure glaucoma assessment in particular.

Group Cost-Sensitive BoostLR With Vector Form Decorrelated Filters for Pedestrian Detection

Pedestrian detection has achieved notable progress in the field of computer vision over the past decade. However, existing top-performing approaches suffer from high computational complexity which prohibits their realization on embedded platforms with low computational capabilities. In this paper, we propose a robust and fast pedestrian detection framework which is based on the Filtered Channel Feature (FCF) approach. The proposed framework exploits vector-form decorrelated filters to extract more discriminative channel features while benefiting from low computational complexity. A novel group cost-sensitive BoostLR (Boosting with Loss Regularization) algorithm is proposed to train the classifier. The proposed training strategy provides more emphasis to the harder samples by exploring the variations of negatives selected from different rounds in hard negative mining processing, and hence is able to boost the overall detection performance. In addition, the proposed method also benefits from the BoostLR framework to achieve better generalization. Experiments on the well-known Caltech, INRIA and CityPersons pedestrian detection datasets show that our proposed approach achieves the best detection performance among all of the state-of-the-art non-deep learning methods and can run one order of magnitude faster than classical FCF methods (e.g. Checkerboards).

IDRiD: Diabetic Retinopathy - Segmentation and Grading Challenge.

Diabetic Retinopathy (DR) is the most common cause of avoidable vision loss, predominantly affecting the working-age population across the globe. Screening for DR, coupled with timely consultation and treatment, is a globally trusted policy to avoid vision loss. However, implementation of DR screening programs is challenging due to the scarcity of medical professionals able to screen a growing global diabetic population at risk for DR. Computer-aided disease diagnosis in retinal image analysis could provide a sustainable approach for such large-scale screening effort. The recent scientific advances in computing capacity and machine learning approaches provide an avenue for biomedical scientists to reach this goal. Aiming to advance the state-of-the-art in automatic DR diagnosis, a grand challenge on "Diabetic Retinopathy - Segmentation and Grading" was organized in conjunction with the IEEE International Symposium on Biomedical Imaging (ISBI - 2018). In this paper, we report the set-up and results of this challenge that is primarily based on Indian Diabetic Retinopathy Image Dataset (IDRiD). There were three principal sub-challenges: lesion segmentation, disease severity grading, and localization of retinal landmarks and segmentation. These multiple tasks in this challenge allow to test the generalizability of algorithms, and this is what makes it different from existing ones. It received a positive response from the scientific community with 148 submissions from 495 registrations effectively entered in this challenge. This paper outlines the challenge, its organization, the dataset used, evaluation methods and results of top-performing participating solutions. The top-performing approaches utilized a blend of clinical information, data augmentation, and an ensemble of models. These findings have the potential to enable new developments in retinal image analysis and image-based DR screening in particular.

Bucketed common vector scaling for authorship attribution in heterogeneous web collections: A scaling approach for authorship attribution

Domain, genre and topic influences on author style adversely affect the performance of authorship attribution (AA) in multi-genre and multi-domain data sets. Although recent approaches to AA tasks focus on suggesting new feature sets and sampling techniques to improve the robustness of a classification system, they do not incorporate domain-specific properties to reduce the negative impact of irrelevant features on AA. This study presents a novel scaling approach, namely, bucketed common vector scaling, to efficiently reduce negative domain influence without reducing the dimensionality of existing features; therefore, this approach is easily transferable and applicable in a classification system. Classification performances on English-language competition data sets consisting of emails and articles and Turkish-language web documents consisting of blogs, articles and tweets indicate that our approach is very competitive to top-performing approaches in English competition data sets and is significantly improving the top classification performance in mixed-domain experiments on blogs, articles and tweets.

LIDAR–camera fusion for road detection using fully convolutional neural networks

In this work, a deep learning approach has been developed to carry out road detection by fusing LIDAR point clouds and camera images. An unstructured and sparse point cloud is first projected onto the camera image plane and then upsampled to obtain a set of dense 2D images encoding spatial information. Several fully convolutional neural networks (FCNs) are then trained to carry out road detection, either by using data from a single sensor, or by using three fusion strategies: early, late, and the newly proposed cross fusion. Whereas in the former two fusion approaches, the integration of multimodal information is carried out at a predefined depth level, the cross fusion FCN is designed to directly learn from data where to integrate information; this is accomplished by using trainable cross connections between the LIDAR and the camera processing branches.To further highlight the benefits of using a multimodal system for road detection, a data set consisting of visually challenging scenes was extracted from driving sequences of the KITTI raw data set. It was then demonstrated that, as expected, a purely camera-based FCN severely underperforms on this data set. A multimodal system, on the other hand, is still able to provide high accuracy. Finally, the proposed cross fusion FCN was evaluated on the KITTI road benchmark where it achieved excellent performance, with a MaxF score of 96.03%, ranking it among the top-performing approaches.

A community effort to assess and improve drug sensitivity prediction algorithms.

Predicting the best treatment strategy from genomic information is a core goal of precision medicine. Here we focus on predicting drug response based on a cohort of genomic, epigenomic and proteomic profiling data sets measured in human breast cancer cell lines. Through a collaborative effort between the National Cancer Institute (NCI) and the Dialogue on Reverse Engineering Assessment and Methods (DREAM) project, we analyzed a total of 44 drug sensitivity prediction algorithms. The top-performing approaches modeled nonlinear relationships and incorporated biological pathway information. We found that gene expression microarrays consistently provided the best predictive power of the individual profiling data sets; however, performance was increased by including multiple, independent data sets. We discuss the innovations underlying the top-performing methodology, Bayesian multitask MKL, and we provide detailed descriptions of all methods. This study establishes benchmarks for drug sensitivity prediction and identifies approaches that can be leveraged for the development of new methods.