Advancements in Multiple Sclerosis Disease Classification Through Machine Learning

McAlpine's Multiple Sclerosis

Runtime evaluation of cognitive systems for non-deterministic multiple output classification problems

Perinatal depression (PND) is one of the most common medical complications during pregnancy and postpartum period, affecting 10–20% of pregnant individuals. Black and Latina women have higher rates of PND, yet they are less likely to be diagnosed and receive treatment. Machine learning (ML) models based on Electronic Medical Records (EMRs) have been effective in predicting postpartum depression in middle-class White women but have rarely included sufficient proportions of racial and ethnic minorities, which contributed to biases in ML models for minority women. Our goal is to determine whether ML models could serve to predict depression in early pregnancy in racial/ethnic minority women by leveraging EMR data. We extracted EMRs from a hospital in a large urban city that mostly served low-income Black and Hispanic women (N=5,875) in the U.S. Depressive symptom severity was assessed from a self-reported questionnaire, PHQ-9. We investigated multiple ML classifiers, used Shapley Additive Explanations (SHAP) for model interpretation, and determined model prediction bias with two metrics, Disparate Impact, and Equal Opportunity Difference. While ML model (Elastic Net) performance was low (ROCAUC=0.67), we identified well-known factors associated with PND, such as unplanned pregnancy and being single, as well as underexplored factors, such as self-report pain levels, lower levels of prenatal vitamin supplement intake, asthma, carrying a male fetus, and lower platelet levels blood. Our findings showed that despite being based on a sample mostly composed of 75% low-income minority women (54% Black and 27% Latina), the model performance was lower for these communities. In conclusion, ML models based on EMRs could moderately predict depression in early pregnancy, but their performance is biased against low-income minority women.

In order to accurately predict soil properties, various machine learning (ML) approaches and hybrid models constructed by integrating ML into regression kriging framework were used to predict and map arable land topsoil pH in Henan province, central China. Random forest (RF), cubist (Cu), support vector machine, artificial neural network, multiple linear regression, classification and regression trees (CART) and their hybrid models were compared for pH accuracy prediction. Among all standalone ML models, RF had the best predictive performance, in terms of the metrics employed in this study, followed by Cu, and CART was the worst. Compared with their ML counterparts, hybrid models could improve the accuracy of topsoil pH prediction to various extents. The accuracy improvement of the hybrid models constructed based on the simple ML was much greater than that based on the complex ensemble ML. Except for artificial neural network kriging , there was no significant difference between different hybrid models in the predicted results of topsoil pH. The outputs from the best predictive models showed that weak acidic soils and weak alkaline soils were the dominant arable soils in the study region, accounting for more than 30% and more than 50% of the total arable land area respectively, the topsoil pH of arable land in the north of the study area is generally higher than that in the south. Boruta variable selection revealed that altitude, climatic covariates closely related to soil moisture availability and some soil properties were the most critical factors affecting and controlling the topsoil pH of arable land.

In medical filed, predicting the occurrence of heart diseases is a significant piece of work. Millions of healthcare-related complexities that have remained unsolved up until now can be greatly simplified with the help of machine learning. The proposed study is concerned with the cardiac disease diagnosis decision support system. An OpenML repository data stream with 1 million instances of heart disease and 14 features is used for this study. After applying to preprocess and feature engineering techniques, machine learning approaches like random forest, decision trees, gradient boosted trees, linear support vector classifier, logistic regression, one-vs-rest, and multilayer perceptron are used to perform binary and multiclassification on the data stream. When combined with the Max Abs Scaler technique, the multilayer perceptron performed satisfactorily in both binary (Accuracy 94.8%) and multiclassification (accuracy 88.2%). Compared to the other binary classification algorithms, the GBT delivered the right outcome (accuracy of 95.8%). Multilayer perceptrons, however, did well in multiple classifications. Techniques such as oversampling and undersampling have a negative impact on disease prediction. Machine learning methods like multilayer perceptrons and ensembles can be helpful for diagnosing cardiac conditions. For this kind of unbalanced data stream, sampling techniques like oversampling and undersampling are not practical.

Lung cancer posed a significant global health challenge, necessitating innovative approaches for early detection and accurate diagnosis. In this paper, CT scan images for lung cancer with three classes namely benign, malignant, and normal are collected from Kaggle. We initially applied conventional machine learning (ML) algorithms including support vector machine (SVM), random forests (RF), decision trees (DT), logistic regression (LR), naive bayes (NB), and k-nearest neighbor for lung cancer detection. The results with these conventional algorithms are recorded. Later, we proposed a novel hybrid model that integrated diverse machine learning algorithms to further enhance accuracy. Our approach combined the power of autoencoders for feature extraction. Using Autoencoder technique, features from images are extracted and a new feature vector is created. Later, the same conventional ML classifiers applied and achieved enhanced performance. The hybrid model demonstrated remarkable performance in identifying lung cancer cases when compared to individual classifiers. Through extensive experimentation, we showcased the efficacy of our integrated framework, achieving high accuracy, precision, recall and F1-score metrics across multiple classifiers. This hybrid approach represented a significant advancement in lung cancer detection, offering a versatile and robust solution for early diagnosis and personalized treatment strategies in clinical settings.

Impaired Sequential but Preserved Motor Memory Consolidation in Multiple Sclerosis Disease

Genetic disease is common in the Level IV Neonatal Intensive Care Unit (NICU), but neonatology providers are not always able to identify the need for genetic evaluation. We trained a machine learning (ML) algorithm to predict the need for genetic testing within the first 18 months of life using health record phenotypes. For a decade of NICU patients, we extracted Human Phenotype Ontology (HPO) terms from clinical text with Natural Language Processing tools. Considering multiple feature sets, classifier architectures, and hyperparameters, we selected a classifier and made predictions on a validation cohort of 2,241 Level IV NICU admits born 2020-2021. Our classifier had ROC AUC of 0.87 and PR AUC of 0.73 when making predictions during the first week in the Level IV NICU. We simulated testing policies under which subjects begin testing at the time of first ML prediction, estimating diagnostic odyssey length both with and without the additional benefit of pursuing rGS at this time. Just by using ML to accelerate initial genetic testing (without changing the tests ordered), the median time to first genetic test dropped from 10 days to 1 day, and the number of diagnostic odysseys resolved within 14 days of NICU admission increased by a factor of 1.8. By additionally requiring rGS at the time of positive ML prediction, the number of diagnostic odysseys resolved within 14 days was 3.8 times higher than the baseline. ML predictions of genetic testing need, together with the application of the right rapid testing modality, can help providers accelerate genetics evaluation and bring about earlier and better outcomes for patients.

Brain imaging characteristics of MOG antibody disease are largely unknown and it is unclear whether they differ from those of multiple sclerosis and AQP4 antibody disease. The aim of this study was to identify brain imaging discriminators between those three inflammatory central nervous system diseases in adults and children to support diagnostic decisions, drive antibody testing and generate disease mechanism hypotheses. Clinical brain scans of 83 patients with brain lesions (67 in the training and 16 in the validation cohort, 65 adults and 18 children) with MOG antibody (n = 26), AQP4 antibody disease (n = 26) and multiple sclerosis (n = 31) recruited from Oxford neuromyelitis optica and multiple sclerosis clinical services were retrospectively and anonymously scored on a set of 29 predefined magnetic resonance imaging features by two independent raters. Principal component analysis was used to perform an overview of patients without a priori knowledge of the diagnosis. Orthogonal partial least squares discriminant analysis was used to build models separating diagnostic groups and identify best classifiers, which were then tested on an independent cohort set. Adults and children with MOG antibody disease frequently had fluffy brainstem lesions, often located in pons and/or adjacent to fourth ventricle. Children across all conditions showed more frequent bilateral, large, brainstem and deep grey matter lesions. MOG antibody disease spontaneously separated from multiple sclerosis but overlapped with AQP4 antibody disease. Multiple sclerosis was discriminated from MOG antibody disease and from AQP4 antibody disease with high predictive values, while MOG antibody disease could not be accurately discriminated from AQP4 antibody disease. Best classifiers between MOG antibody disease and multiple sclerosis were similar in adults and children, and included ovoid lesions adjacent to the body of lateral ventricles, Dawson's fingers, T1 hypointense lesions (multiple sclerosis), fluffy lesions and three lesions or less (MOG antibody). In the validation cohort patients with antibody-mediated conditions were differentiated from multiple sclerosis with high accuracy. Both antibody-mediated conditions can be clearly separated from multiple sclerosis on conventional brain imaging, both in adults and children. The overlap between MOG antibody oligodendrocytopathy and AQP4 antibody astrocytopathy suggests that the primary immune target is not the main substrate for brain lesion characteristics. This is also supported by the clear distinction between multiple sclerosis and MOG antibody disease both considered primary demyelinating conditions. We identify discriminatory features, which may be useful in classifying atypical multiple sclerosis, seronegative neuromyelitis optica spectrum disorders and relapsing acute disseminated encephalomyelitis, and characterizing cohorts for antibody discovery.

Medical data is becoming increasingly complex, which highlights the need for automated detection systems.In this paper, a system is proposed that utilizes both machine learning and deep learning techniques to accurately detect multiple diseases.The system makes use of a combination of a convolutional neural network (CNN) and a support vector machine (SVM) to train and classify medical data.To detect different diseases, the pre-trained CNN model is fine-tuned, utilizing transfer learning.The proposed system was evaluated on a dataset of medical images, and it achieved an impressive overall accuracy of 95%.This system has the potential to aid medical practitioners in the early detection and diagnosis of multiple diseases. I.

Robust method for classification of agricultural crops diseases using LGXP descriptor

Multiple Sclerosis (MS) is a chronic immune-mediated inflammatory disease (IMID) of the central nervous system (CNS). Early identification of MS, especially as a screening method for at-risk individuals, is crucial to delay disease progression and improve patient outcomes by preventing future irreversible neurologic damage. In this work, we utilize well-validated tracking scanning laser ophthalmoscope (TSLO) image to predict MS compared to the unaffected controls. While traditional Machine Learning (ML) methods, such as Logistic Regression (LR), have demonstrated a strong predictive power [Mauro F. Pinto et al., 2020] in disease identification, we propose the use of a novel DL based model. Though the use of Deep Neural Network (DNN), this model can have a much higher learning capacity to capture latent features embedded in the retinal images. We hypothesize that such latent information, often hidden in ML feature engineering processes, plays an important role for the prediction of disease and can be well represented by DL models. To establish a DL based model capable of learning latent image features to provide predictive power for the presence of MS. Utilize a deep convolutional neural network to extract the retinal coding and implement a recurrent neural network to learn the temporal correlations in video sequences. Our approaches were tested using a 250-subject MS/control database collected at the UCSF. Patients with Expanded Disability Status Scale (EDSS)< 4 are compared to healthy subjects. Both raw retinal images and the frequency and spatial patterns of the eye motion are combined to construct a hybrid image, denoted as "retinal coding", and directly fed to the DL model for training and testing. Preliminary results on predictive power were measured using Area-under-Curve (AUC) of the Receiver Operating Characteristics (ROC) curve, sensitivity, and specificity, as well as an F-1 score. We observe an AUC of 0.920, sensitivity of 0.90, specificity of 0.89 and F-1 score of 0.89 using the DL model to distinguish MS from controls, which outperforms the baseline LR model by 24%. This work can be considered as a proof-of-concept concerning the possibility of identifying MS disease using a DL based approach. The results demonstrate the possibility of predicting early-stage MS and understanding disease's dynamics. Such end-to-end model could be generalizable and trained on other disease states.

Predicting stock returns by classifier ensembles

Today, network-based intrusions are among the most prevalent security threats our networked systems face. In the case of software-defined networks (SDN), not only the connected devices and services but also the SDN controllers may be subjected to intrusion attempts. The advent of efficient and robust machine learning (ML) algorithms along with the availability of a large number of network datasets enabled the development of ML-based network intrusion detection systems (NIDS). Recent work has demonstrated that ML-based NIDS systems are vulnerable to evasion attacks where the adversary targets the ML classifier in the NIDS system to evade detection by performing various packet perturbations. In this work, we propose an approach to build robust ML based NIDS systems that use multiple ML classifiers trained with reduced feature sets. Our approach depends on a careful feature selection procedure based on Permutation Feature Importance, a wrapper based feature engineering method. Our evaluations on well-known datasets show that the proposed hybrid multi-classifier system is robust and performs well against the packet perturbation attacks considered in this work.

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition having symptoms such as difficulties in social communication, restricted interest and repetitive behaviours. It is crucial to predict ASD at an early stage, as this can improve long-term developmental outcomes. The drawbacks of existing ASD diagnostic approaches are that they are timeconsuming, subjective and rely on clinical expertise. To overcome these drawbacks, this study explores the use of machine learning algorithm for early prediction of ASD. We use multiple classifiers namely logistic regression, support vector machines, random forest, k-nearest neighbor and artificial neural networks. These models are trained and evaluated on an ASD dataset from the UCI Machine Learning repository. To make sure a robust and unbiased performance assessment, stratified <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{k}$</tex>-fold cross validation is used. This maintains class distribution across various folds and avoids the risks of overfitting. Experimental results show that machine learning models, especially artificial neural networks, achieve good accuracy, precision, recall, F1-score and AUC-ROC scores. This study demonstrates the potential of machine learning algorithms for the early prediction of ASD which is reliable and cost-effective.