Binary Classification Research Articles

Symmetry-induced ambiguity in orientation estimation from RGB images

The estimation of object orientation from RGB images is a core component in many modern computer vision pipelines. Traditional techniques mostly predict a single orientation per image, learning a one-to-one mapping between images and rotations. However, when objects exhibit rotational symmetries, they can appear identical from multiple viewpoints. This induces ambiguity in the estimation problem, making images map to rotations in a one-to-many fashion. In this paper, we explore several ways of addressing this problem. In doing so, we specifically consider algorithms that can map an image to a range of multiple rotation estimates, accounting for symmetry-induced ambiguity. Our contributions are threefold. Firstly, we create a data set with annotated symmetry information that covers symmetries induced through self-occlusion. Secondly, we compare and evaluate various learning strategies for multiple-hypothesis prediction models applied to orientation estimation. Finally, we propose to model orientation estimation as a binary classification problem. To this end, based on existing work from the field of shape reconstruction, we design a neural network that can be sampled to reconstruct the full range of ambiguous rotations for a given image. Quantitative evaluation on our annotated data set demonstrates its performance and motivates our design choices.

Rethinking Gender: Beyond the Binary and into the Unknown.

Our present and evolved understanding has challenged the previously synonymous use of the terms 'sex' and 'gender'. We have moved beyond the binary categorization towards proliferation of gender identities. Thus, raising questions whether certain identities are traits or gender identities. This complexity of the issue is exacerbated by the cultural relativity of gender identity and by lack of a standardized list. Adopting a balanced approach, the article touches upon the prejudices against the gender minorities. Additionally, it touches upon the controversies surrounding gender identities and their development to ensure that the individuals can make more informed choices and engage in more meaningful discourse. It addresses issues of politicised debates, linguistic diversity, and the role of pharma industry in the sex-change procedures. The paradigm of gender has transcended the binary constructs in the contemporary discourse. However, it has ventured into unchartered territories revealing several unexplored facets that await scholarly investigation. The present paper critiques the concept of gender identities and the sociopolitical landscape surrounding it through the lens of Critical Theory. In conclusion, our understanding of gender is still limited and evolving. There is a need for adopting a more nuanced and informed approach to challenge the issues posed by this era of evolving gender expression and identities. The article concludes with policy recommendations based on insights gained from the article and suggestions for future research.

Multiclass Synthetic Accessibility Prediction.

Evaluating synthetic accessibility of in silico molecules is an integral component of the drug discovery process. While the application of machine learning models to predict whether small molecules are easy or hard to synthesize has gained attention recently, predetermined thresholds and data set imbalances present challenges for these binary classification approaches. In this study, we introduce a novel multiclass fold-ensembled classification approach to predict the minimum number of steps needed to synthesize a small molecule. By ensembling the base models trained on multiple stratified subsampled folds, this approach effectively mitigates the impact of class imbalance through probability aggregation or voting aggregation strategies. Additionally, we propose fuzzy evaluation metrics that account for practical tolerances in predictions, providing a more flexible and realistic assessment of model performance. Through experimentation on two reaction benchmark data sets, we demonstrate the effectiveness of our model in a multiclass synthetic accessibility prediction task and the superiority of our proposed method over six existing models in binary synthetic accessibility prediction tasks.

Deep learning-based real-time detection and classification of tomato ripeness stages using YOLOv8 on raspberry Pi

Abstract The automated detection of tomato ripeness is critical in crop management and harvesting. In most earlier works, tomato image ripeness detection has been based upon a limited set of images and binary classification (ripe and unripe). This study uses the cutting-edge YOLOv8 object detection algorithm and a comprehensive dataset to propose an accurate real-time system for detecting and classifying tomato ripeness (multi-class). Based on two open-source datasets (Kaggle and Internet-sourced), we developed and tested the proposed system. In this method, a comprehensive tomato image dataset is curated, YOLOv8 models are built, seamlessly integrated into an embedded system (Raspberry Pi4), then evaluated and validated. The model shows exceptional performance in detecting three distinct classes of ripeness: green, partially ripe, and ripe. It surpasses existing state-of-the-art models in both accuracy and efficiency. Based on the Kaggle dataset, our model achieves an average precision at 50 of 0.808, with F1-scores of 0.80, 0.65, and 0.796 for green, partially ripe, and ripe classes, respectively. It achieves mAP at 50 of 0.725 and F1-scores of 0.747 (green), 0.652 (partially ripe), and 0.72 (ripe) for the corresponding classes of the Internet-sourced Dataset, exceeding current state-of-the-art models. Finally, the proposed tomato ripeness detection algorithm is implemented on the Raspberry Pi4 system and exhibits notable performance. With the integration of YOLOv8 into an embedded system (Respbeery Pi4), it can be used to improve efficiency and reduce labor costs in tomato-picking robots, helping to revolutionize agricultural practices.

Multiclass hate speech detection with an aggregated dataset

Abstract Detecting and removing hate speech content in a timely manner remains a challenge for social media platforms. Automated techniques such as deep learning models offer solutions which can keep up with the volume and velocity of user content production. Research in this area has mainly focused on either binary classification or on classifying tweets into generalised categories such as hateful, offensive, or neither. Less attention has been given to multiclass classification of online hate speech into the type of hate or group at which it is directed. By aggregating and re-annotating several relevant hate speech datasets, this study presents a dataset and evaluates several models for classifying tweets into the categories ethnicity, gender, religion, sexuality, and non-hate. We evaluate the dataset by training several models: logistic regression, LSTM, BERT, and GPT-2. For the LSTM model, we assess a range of NLP features using a multi-classification LSTM model, and conclude that the highest performing feature combination consists of word $n$ -grams, character $n$ -grams, and dependency tuples. We show that while more recent larger models can achieve a slightly higher performance, increased model complexity alone is not sufficient to achieve significantly improved models. We also compare this approach with a binary classification approach and evaluate the effect of dataset size on model performance.

Comparative analysis of deep learning models for crack detection in buildings

Life-time of the buildings is generally challenged by the act of nature. In-spite of the fact that the constructions provide minimum guarantee on quality and durability, certain mismatch in the composition of the materials, stress on the building, and chemical or physical imbalance of the materials, lead to surface crack. Cracks are also generated due to the shuffle of climatic conditions, which leads to the contraction and expansion of the building surfaces, and other damages. The guarantee on building safety and serviceability depends on how these buildings are successfully assessed and maintained. The development of Artificial Intelligence (AI) techniques, provide favourable solutions in-order to handle, manage and solve building cracks, through analysis using deep image neural network models, that perform classification of the building with crack images. As a result, a critical challenge for many civil engineering applications is the precise, quick, and automated identification of cracks on structural surfaces is addressed with the solutions provided by the deep image neural networks. In this research, we tackle the research gap and data scarcity by developing and curating a novel deep learning image processing for detecting cracks in brickwork. We also train and validate several deep learning models to classify brickwork images as either cracked or normal. The dataset of the proposed work contains 24,000 images which are classified through binary classes. These classes are generated for crack and non-crack images. The various parameters such as Batch size, Pooling, Activation functions Learning-rate, Kernel-Size, Normalization, and Optimizers are used for the evaluation of the model. The proposed work performs a comparative analysis of four deep image models such as Inception V3, VGG-16, RESNET-50 VGG-19, Inception ResNetV2 and CNN-RES MLP. With the analysis of all these models, the Inception V3 provides the best of all with the accuracy value of 99.98%. The InceptionV3 tops the Precision value of 99.99% and RESNET-50 tops the Recall value of 99.98%. The IncpetionV2 provided the best of the Region of Convergence value of 0.9999 which is the best among all the models for reliable and stable performance.

Enhancing supply chain resilience through supervised machine learning: supplier performance analysis and risk profiling for a multi-class classification problem

PurposeThis paper explores the application of supervised machine learning (ML) classification models to address supplier performance analysis and risk profiling as a multi-class classification problem. The research highlights that current applications of machine learning in supplier selection primarily focus on binary classification problems, underscoring a significant gap in the literature.Design/methodology/approachThis research paper opts for a structured approach to solve supplier selection and risk profiling using supervised machine learning multi-class classification models and prediction probabilities. The study involved a synthetic data set of 1,600 historical data points, creating a supplier selection framework that simulates current supply chain (SC) performance. The “Supplier Analysis and Selection ML Module” guided supplier selection recommendations based on ML analysis. Real-world variability is introduced through random seeds, impacting actual delivery dates, quantity delivered and quality performance. Supervised ML models, with hyperparameter tuning, enable multi-class classification of suppliers, considering past delivery performance and risk calculations.FindingsThe study demonstrates the effectiveness of the supervised ML-based approach in ensuring consistent supplier selection across multi-class classification problems. Beyond evaluating past delivery performance, it introduces a new dimension by predicting and assessing supplier risks through ML-generated prediction probabilities. This can enhance overall SC visibility and help organizations optimize strategies associated with risk mitigation, inventory management and customer service.Research limitations/implicationsThe findings highlight the adaptability of ML-based methodologies in dynamic SC environments, providing a proactive means to identify and manage risks. These insights are vital for organizations aiming to bolster SC resilience, particularly amid uncertainties.Practical implicationsThe practical implications of this study are significant for both commercial and humanitarian supply chain management (SCM). For commercial applications, the ML-based methodology allows businesses to make more informed supplier selection decisions, reducing risks and improving operational efficiency. In disaster and humanitarian SC contexts, the use of ML can improve preparedness and resource allocation, ensuring that critical supplies reach affected areas promptly.Social implicationsThe study’s implications extend to disaster and humanitarian SCM, where timely and efficient delivery is critical for saving lives and alleviating suffering. ML tools can improve preparedness, resource allocation and coordination in these contexts, enhancing the resilience and responsiveness of humanitarian supply chains.Originality/valueUnlike conventional methods focused on quality, cost and delivery performance aspects, the current study introduces supervised ML to identify and assess supplier risks through prediction probabilities for multi-class classification problems (delivery performance as late, on-time and ahead), offering a refined understanding of supplier selection in dynamic SC environments.

Performance analysis of capsule networks for detecting GPS spoofing attacks on unmanned aerial vehicles

Unmanned aerial vehicles (UAVs) are prone to several cyber-attacks, including global positioning system (GPS) spoofing. The use of machine learning and deep learning are becoming increasingly common for UAV GPS spoofing attack detection; however, these approaches have some limitations, such as a high rate of false alarm and misdetection. We propose using capsule networks to detect and classify UAV-focused GPS spoofing attacks. This paper compares simple capsule networks, efficient capsule networks, dual attention capsule networks, and convolutional neural network in terms of accuracy, probability of detection, probability of misdetection, probability of false alarm, prediction time, training time per sample, and memory size. The results indicate that the Efficient-capsule network outperforms the other models, as demonstrated by an accuracy of 99.1%, a probability of detection of 99.9%, a probability of misdetection of 0.1%, a probability of false alarm of 0.37%, a prediction time of 0.5 seconds, a training time per sample of 0.2 seconds, and a memory size of 123 mebibytes for binary classification.

Assessing juicy elements in interactive infographics

Introduction: Juiciness refers to the use of various audiovisual effects that are triggered in response to user interactions. This study explores the presence of juicy elements in interactive infographics, thereby extending the investigation of juiciness beyond traditional gaming contexts. Methodology: A descriptive research approach, employing content analysis, was used to evaluate a sample of interactive infographics published between 2010 and 2024. The sample was collected from four prominent sources, and each visualization was analyzed using a binary classification system (YES or NO) to indicate the presence of the identified juiciness elements: animation, particles, audio feedback, screen shake, and persistence. Results: Overall, the findings indicate that juicy elements are present in certain interactive visualizations, though not universally across all examples. The data revealed that animation appeared most frequently, with an occurrence rate of 73.85%, followed by particles (20.51%), audio feedback (5.64%), persistence (4.1%), and screen shake (1.03%). Furthermore, 25.64% of the visualizations contained no juicy elements. Discussion: The findings from the analysis reveal that juicy elements are present in a significant number of interactive visualizations, but none of the visualizations analyzed incorporated all five juicy elements simultaneously. Despite the presence of some juicy elements, no single visualization captured the full essence of juicy design, which ideally offers a high level of feedback from minimal user input. Conclusions: While no single visualization incorporated all five juicy elements, combinations of up to four were observed, suggesting that juiciness does not require a uniform or exhaustive application of all elements.

Sentiment Analysis Techniques for Deep Learning Classification and Comparison

Sentiment analysis, a critical tool in natural language processing (NLP), has seen significant advancements with the rise of machine learning and deep learning techniques. This paper explores the evolution of sentiment analysis from traditional models, such as CNN and RNN, to more sophisticated approaches like LSTM, LSTM-CNN, and BERT. Using the IMDb dataset for binary classification of sentiment, we utilize metrics like F-Score, Precision, Accuracy, and Recall to compare the way these models perform. Despite the progress made, sentiment analysis still faces challenges, particularly in handling multiple languages, incorporating multimedia content, and addressing privacy and ethical concerns. This paper also discusses future directions for sentiment analysis, including the development of models for low-resource languages, real-time sentiment analysis, and the ethical implications of data handling. Overall, the study highlights the potential of sentiment analysis and the opportunities for improvement through continued research and innovation.

Characterization of adrenal glands on computed tomography with a 3D V-Net-based model

ObjectivesTo evaluate the performance of a 3D V-Net-based segmentation model of adrenal lesions in characterizing adrenal glands as normal or abnormal.MethodsA total of 1086 CT image series with focal adrenal lesions were retrospectively collected, annotated, and used for the training of the adrenal lesion segmentation model. The dice similarity coefficient (DSC) of the test set was used to evaluate the segmentation performance. The other cohort, consisting of 959 patients with pathologically confirmed adrenal lesions (external validation dataset 1), was included for validation of the classification performance of this model. Then, another consecutive cohort of patients with a history of malignancy (N = 479) was used for validation in the screening population (external validation dataset 2). Parameters of sensitivity, accuracy, etc., were used, and the performance of the model was compared to the radiology report in these validation scenes.ResultsThe DSC of the test set of the segmentation model was 0.900 (0.810–0.965) (median (interquartile range)). The model showed sensitivities and accuracies of 99.7%, 98.3% and 87.2%, 62.2% in external validation datasets 1 and 2, respectively. It showed no significant difference comparing to radiology reports in external validation datasets 1 and lesion-containing groups of external validation datasets 2 (p = 1.000 and p > 0.05, respectively).ConclusionThe 3D V-Net-based segmentation model of adrenal lesions can be used for the binary classification of adrenal glands.Critical relevance statementA 3D V-Net-based segmentation model of adrenal lesions can be used for the detection of abnormalities of adrenal glands, with a high accuracy in the pre-surgical scene as well as a high sensitivity in the screening scene.Key PointsAdrenal lesions may be prone to inter-observer variability in routine diagnostic workflow.The study developed a 3D V-Net-based segmentation model of adrenal lesions with DSC 0.900 in the test set.The model showed high sensitivity and accuracy of abnormalities detection in different scenes.Graphical

Diagnosis of Cognitive and Mental Disorders: A New Approach Based on Spectral–Spatiotemporal Analysis and Local Graph Structures of Electroencephalogram Signals

Background/Objectives: The classification of psychological disorders has gained significant importance due to recent advancements in signal processing techniques. Traditionally, research in this domain has focused primarily on binary classifications of disorders. This study aims to classify five distinct states, including one control group and four categories of psychological disorders. Methods: Our investigation will utilize algorithms based on Granger causality and local graph structures to improve classification accuracy. Feature extraction from connectivity matrices was performed using local structure graphs. The extracted features were subsequently classified employing K-Nearest Neighbors (KNN), Support Vector Machine (SVM), AdaBoost, and Naïve Bayes classifiers. Results: The KNN classifier demonstrated the highest accuracy in the gamma band for the depression category, achieving an accuracy of 89.36%, a sensitivity of 89.57%, an F1 score of 94.30%, and a precision of 99.90%. Furthermore, the SVM classifier surpassed the other machine learning algorithms when all features were integrated, attaining an accuracy of 89.06%, a sensitivity of 88.97%, an F1 score of 94.16%, and a precision of 100% for the discrimination of depression in the gamma band. Conclusions: The proposed methodology provides a novel approach for analyzing EEG signals and holds potential applications in the classification of psychological disorders.

Scoring of swine lung images: a comparison between a computer vision system and human evaluators

Cranioventral pulmonary consolidation (CVPC) is a common lesion observed in the lungs of slaughtered pigs, often associated with Mycoplasma (M.) hyopneumoniae infection. There is a need to implement simple, fast, and valid CVPC scoring methods. Therefore, this study aimed to compare CVPC scores provided by a computer vision system (CVS; AI DIAGNOS) from lung images obtained at slaughter, with scores assigned by human evaluators. In addition, intra- and inter-evaluator variability were assessed and compared to intra-CVS variability. A total of 1050 dorsal view images of swine lungs were analyzed. Total lung lesion score, lesion score per lung lobe, and percentage of affected lung area were employed as outcomes for the evaluation. The CVS showed moderate accuracy (62–71%) in discriminating between non-lesioned and lesioned lung lobes in all but the diaphragmatic lobes. A low multiclass classification accuracy at the lung lobe level (24–36%) was observed. A moderate to high inter-evaluator variability was noticed depending on the lung lobe, as shown by the intraclass correlation coefficient (ICC: 0.29–0.6). The intra-evaluator variability was low and similar among the different outcomes and lung lobes, although the observed ICC slightly differed among evaluators. In contrast, the CVS scoring was identical per lobe per image. The results of this study suggest that the CVS AI DIAGNOS could be used as an alternative to the manual scoring of CVPC during slaughter inspections due to its accuracy in binary classification and its perfect consistency in the scoring.

All-optical perception based on partially coherent optical neural networks

In the field of image processing, optical neural networks offer advantages such as high speed, high throughput, and low energy consumption. However, most existing coherent optical neural networks (CONN) rely on coherent light sources to establish transmission models. The use of laser inputs and electro-optic modulation devices at the front end of these neural networks diminishes their computational capability and energy efficiency, thereby limiting their practical applications in object detection tasks. This paper proposes a partially coherent optical neural network (PCONN) transmission model based on mutual intensity modulation. This model does not depend on coherent light source inputs or active electro-optic modulation devices, allowing it to directly compute and infer using natural light after simple filtering, thus achieving full optical perception from light signal acquisition to computation and inference. Simulation results indicate that the model achieves a highest classification accuracy of 96.80% and 86.77% on the MNIST and Fashion-MNIST datasets, respectively. In a binary classification simulation test based on the ISDD segmentation dataset, the model attained an accuracy of 94.69%. It is estimated that this system’s computational inference speed for object detection tasks is 100 times faster than that of traditional CONN, with energy efficiency approximately 50 times greater. In summary, our proposed PCONN model addresses the limitations of conventional optical neural networks in coherent light environments and is anticipated to find applications in practical object detection scenarios.

Des-q: a quantum algorithm to provably speedup retraining of decision trees

Decision trees are widely adopted machine learning models due to their simplicity and explainability. However, as training data size grows, standard methods become increasingly slow, scaling polynomially with the number of training examples. In this work, we introduce Des-q, a novel quantum algorithm to construct and retrain decision trees for regression and binary classification tasks. Assuming the data stream produces small, periodic increments of new training examples, Des-q significantly reduces the tree retraining time. Des-q achieves a logarithmic complexity in the combined total number of old and new examples, even accounting for the time needed to load the new samples into quantum-accessible memory. Our approach to grow the tree from any given node involves performing piecewise linear splits to generate multiple hyperplanes, thus partitioning the input feature space into distinct regions. To determine the suitable anchor points for these splits, we develop an efficient quantum-supervised clustering method, building upon the q-means algorithm introduced by Kerenidis et al. We benchmark the simulated version of Des-q against the state-of-the-art classical methods on multiple data sets and observe that our algorithm exhibits similar performance to the state-of-the-art decision trees while significantly speeding up the periodic tree retraining.

Automated classification of coronary LEsions fRom coronary computed Tomography angiography scans with an updated deep learning model: ALERT study.

The use of deep learning models for quantitative measurements on coronary computed tomography angiography (CCTA) may reduce inter-reader variability and increase efficiency in clinical reporting. This study aimed to investigate the diagnostic performance of a recently updated deep learning model (CorEx-2.0) for quantifying coronary stenosis, compared separately with two expert CCTA readers as references. This single-center retrospective study included 50 patients that underwent CCTA to rule out obstructive coronary artery disease between 2017-2022. Two expert CCTA readers and CorEx-2.0 independently assessed all 150 vessels using Coronary Artery Disease-Reporting and Data System (CAD-RADS). Inter-reader agreement analysis and diagnostic performance of CorEx-2.0, compared with each expert reader as references, were evaluated using percent agreement, Cohen's kappa for the binary CAD-RADS classification (CAD-RADS 0-3 versus 4-5) at patient level, and linearly weighted kappa for the 6-group CAD-RADS classification at vessel level. Overall, 50 patients and 150 vessels were evaluated. Inter-reader agreement using the binary classification at patient level was 91.8% (45/49) with a Cohen's kappa of 0.80. For the 6-group classification at vessel level, inter-reader agreement was 67.6% (100/148) with a linearly weighted kappa of 0.77. CorEx-2.0 showed 100% sensitivity for detecting CAD-RADS ≥ 4 and kappa values of 0.86 versus both readers using the binary classification at patient level. For the 6-group classification at vessel level, CorEx-2.0 demonstrated weighted kappa values of 0.71 versus reader 1 and 0.73 versus reader 2. CorEx-2.0 identified all patients with severe stenosis (CAD-RADS ≥ 4) compared with expert readers and approached expert reader performance at vessel level (weighted kappa > 0.70). Question Can deep learning models improve objectivity in coronary stenosis grading and reporting as coronary CT angiography (CTA) workloads rise? Findings The deep learning model (CorEx-2.0) identified all patients with severe stenoses when compared with expert readers and approached expert reader performance at vessel level. Clinical relevance CorEx-2.0 is a reliable tool for identifying patients with severe stenoses (≥ 70%), underscoring the potential of using this deep learning model to prioritize coronary CTA reading by flagging patients at risk of severe obstructive coronary artery disease.

Multimodal multiview bilinear graph convolutional network for mild cognitive impairment diagnosis.

Mild cognitive impairment (MCI) is a significant predictor of the early progression of Alzheimer's disease, and it can be used as an important indicator of disease progression. However, many existing methods focus mainly on the image itself when processing brain imaging data, ignoring other non-imaging data (e.g., genetic, clinical information, etc.) that may have underlying disease information. In addition, imaging data acquired from different devices may exhibit varying degrees of heterogeneity, potentially resulting in numerous noisy connections during network construction. To address these challenges, this study proposes a Multimodal Multiview Bilinear Graph Convolution (MMBGCN) framework for disease risk prediction. Firstly, grey matter (GM), white matter (WM) and cerebrospinal fluid (CSF) features are extracted from magnetic resonance imaging (MRI), and non-imaging information is combined with the features extracted from MRI to construct a multimodal shared adjacency matrix. The shared adjacency matrix is then used to construct the multiview network so that the effect of potential disease information in the non-imaging information on the model can be considered. Finally, the MRI features extracted by the multiview network are weighted to reduce noise, and then the spatial pattern is restored by bilinear convolution. The features of the recovered spatial patterns are then combined with the genetic information for disease prediction. The proposed method is tested on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. Extensive experiments demonstrate the superior performance of the proposed framework and its ability to outperform other related algorithms. The average classification accuracy in the binary classification task in this study is 89.6%. The experimental results demonstrate that the method proposed in this study facilitates research on MCI diagnosis using multimodal data.

Machine learning identified novel players in lipid metabolism, endosomal trafficking, and iron metabolism of the ALS spinal cord

Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease affecting motor neurons. Although genes causing familial cases have been identified, those of sporadic ALS, which occupies the majority of patients, are still elusive. In this study, we adopted machine learning to build binary classifiers based on the New York Genome Center (NYGC) ALS Consortium’s RNA-seq data of the postmortem spinal cord of ALS and non-neurological disease control. The accuracy of the classifiers was greater than 83% and 77% for the training set and the unseen test set, respectively. The classifiers contained 114 genes. Among them, 41 genes have been reported in previous ALS studies, and others are novel in this field. These genes are involved in mitochondrial respiration, lipid metabolism, endosomal trafficking, and iron metabolism, which may promote the progression of ALS pathology.

A machine learning model accurately identifies glycogen storage disease Ia patients based on plasma acylcarnitine profiles

BackgroundGlycogen storage disease (GSD) Ia is an ultra-rare inherited disorder of carbohydrate metabolism. Patients often present in the first months of life with fasting hypoketotic hypoglycemia and hepatomegaly. The diagnosis of GSD Ia relies on a combination of different biomarkers, mostly routine clinical chemical markers and subsequent genetic confirmation. However, a specific and reliable biomarker is lacking. As GSD Ia patients demonstrate altered lipid metabolism and mitochondrial fatty acid oxidation, we built a machine learning model to identify GSD Ia patients based on plasma acylcarnitine profiles.MethodsWe collected plasma acylcarnitine profiles from 3958 patients, of whom 31 have GSD Ia. Synthetic samples were generated to address the problem of class imbalance in the dataset. We built several machine learning models based on gradient-boosted trees. Our approach included hyperparameter tuning and feature selection and generalization was checked using both nested cross-validation and a held-out test set.ResultsThe binary classifier was able to correctly identify 5/6 GSD Ia patients in a held-out test set without generating significant amounts of false positive results. The best model showed excellent performance with a mean received operator curve (ROC) AUC of 0.955 and precision-recall (PR) curve AUC of 0.674 in nested CV.ConclusionsThis study demonstrates an innovative approach to applying machine learning to ultra-rare diseases by accurately identifying GSD Ia patients based on plasma free carnitine and acylcarnitine concentrations, leveraging subtle acylcarnitine abnormalities. Acylcarnitine features that were strong predictors for GSD Ia include C16-carnitine, C14OH-carnitine, total carnitine and acetylcarnitine. The model demonstrated high sensitivity and specificity, with selected parameters that were not only robust but also highly interpretable. Our approach offers potential prospect for the inclusion of GSD Ia in newborn screening. Rare diseases are underrepresented in machine learning studies and this work highlights the potential for these techniques, even in ultra-rare diseases such as GSD Ia.

Comparing multi-texture fibrosis analysis versus binary opacity-based abnormality detection for quantitative assessment of idiopathic pulmonary fibrosis

Automated tools for quantification of idiopathic pulmonary fibrosis (IPF) can aid in ensuring reproducibility, however their complexity and costs can differ substantially. In this retrospective study, two automated tools were compared in 45 patients with biopsy proven (12/45) and imaging-based (33/45) IPF diagnosis (mean age 74 ± 9 years, 37 male) for quantification of pulmonary fibrosis in CT. First, a tool that identifies multiple characteristic lung texture features was applied to measure multi-texture fibrotic lung (MTFL) by combining the amount of ground glass, reticulation, and honeycombing. Opacity-based fibrotic lung (OFL) was measured by a second tool that performs a simpler binary classification of tissue into either normal or opacified lung and was originally developed for quantifying pneumonia. Differences in quantification of MTFL and OFL were assessed by Mann-Whitney U-test and Pearson correlation (r). Also, correlation with spirometry parameters (percent predicted total lung capacity (TLC), percent predicted vital capacity (VC), percent predicted forced expiratory volume in 1 s (FEV1), diffusing capacity of the lungs for carbon monoxide (DLCO), partial pressure of oxygen (PO2) and carbon dioxide (PCO2)) were assessed by r. The prognostic values for 3-year patient survival of OFL, LSS and MTFL were investigated by multivariable Cox-proportional-hazards (CPH) models including sex, age and TLC and including sex, age and VC. Also, Kaplan-Meier analysis with log rank test between subgroups separated by median OFL and MTFL were conducted. No significant difference between OFL and MTFL was observed (median and interquartile range: OFL = 29% [20-38%], MTFL = 31% [19-45%]; P = 0.44). For OFL significant correlation was observed to MTFL (r = 0.93, P < 0.01) and VC (r=-0.50, P = 0.03). For MTFL no significant correlation to spirometry parameters was found. The total time for one analysis was lower for the automated MTFL (MTFL: 313 ± 25s vs. OFL: 612 ± 61s, P < 0.001). Both analyses were significant predictors in the multivariable CPH analysis including TLC (hazard-ratios: MTFL 1.03 [1.01–1.06], P = 0.02; OFL 1.03 [1.00-1.06], P = 0.03). No parameter was a significant predictor in the CPH models including VC (hazard-ratios: MTFL 1.01 [0.98–1.04], P = 1; OFL 1.01 [0.97–1.05], P = 1). OFL showed significance in Kaplan-Meier analysis (MTFL: P = 0.17; OFL: P = 0.03). Using a simple opacity-based quantification of pulmonary fibrosis in IPF patients displayed similar results and prognostic value compared to a more complex multi-texture based analysis.