Brain age prediction of healthy subjects on anatomic MRI with deep learning: going beyond with an "explainable AI" mindset

Objectives: Define a clinically usable preprocessing pipeline for MRI data. Predict brain age using various machine learning and deep learning algorithms. Define Caveat against common machine learning traps. Data and Methods: We used 1597 open-access T1 weighted MRI from 24 hospitals. Preprocessing consisted in applying: N4 bias field correction, registration to MNI152 space, white and grey stripe intensity normalization, skull stripping and brain tissue segmentation. Prediction of brain age was done with growing complexity of data input (histograms, grey matter from segmented MRI, raw data) and models for training (linear models, non linear model such as gradient boosting over decision trees, and 2D and 3D convolutional neural networks). Work on interpretability consisted in (i) proceeding on basic data visualization like correlations maps between age and voxels value, and generating (ii) weights maps of simpler models, (iii) heatmap from CNNs model with occlusion method. Results: Processing time seemed feasible in a radiological workflow: 5 min for one 3D T1 MRI. We found a significant correlation between age and gray matter volume with a correlation r = -0.74. Our best model obtained a mean absolute error of 3.60 years, with fine tuned convolution neural network (CNN) pretrained on ImageNet. We carefully analyzed and interpreted the center effect. Our work on interpretability on simpler models permitted to observe heterogeneity of prediction depending on brain regions known for being involved in ageing (grey matter, ventricles). Occlusion method of CNN showed the importance of Insula and deep grey matter (thalami, caudate nuclei) in predictions. Conclusions: Predicting the brain age using deep learning could be a standardized metric usable in daily neuroradiological reports. An explainable algorithm gives more confidence and acceptability for its use in practice. More clinical studies using this new quantitative biomarker in neurological diseases will show how to use it at its best.

Download Full-text

Systematic Evaluation of Machine Learning Algorithms for Neuroanatomically-Based Age Prediction in Youth

10.1101/2021.11.24.469888 ◽

2021 ◽

Author(s):

Mandana Modabbernia ◽

Heather C Whalley ◽

David Glahn ◽

Paul M. Thompson ◽

Rene S. Kahn ◽

...

Keyword(s):

Machine Learning ◽

Computational Efficiency ◽

Learning Algorithms ◽

Machine Learning Algorithms ◽

Sensitivity Analyses ◽

Gradient Boosting ◽

Support Vector ◽

Age Related ◽

Extreme Gradient Boosting ◽

Brain Age

Application of machine learning algorithms to structural magnetic resonance imaging (sMRI) data has yielded behaviorally meaningful estimates of the biological age of the brain (brain-age). The choice of the machine learning approach in estimating brain-age in children and adolescents is important because age-related brain changes in these age-groups are dynamic. However, the comparative performance of the multiple machine learning algorithms available has not been systematically appraised. To address this gap, the present study evaluated the accuracy (Mean Absolute Error; MAE) and computational efficiency of 21 machine learning algorithms using sMRI data from 2,105 typically developing individuals aged 5 to 22 years from five cohorts. The trained models were then tested in an independent holdout datasets, comprising 4,078 pre-adolescents (aged 9-10 years). The algorithms encompassed parametric and nonparametric, Bayesian, linear and nonlinear, tree-based, and kernel-based models. Sensitivity analyses were performed for parcellation scheme, number of neuroimaging input features, number of cross-validation folds, and sample size. The best performing algorithms were Extreme Gradient Boosting (MAE of 1.25 years for females and 1.57 years for males), Random Forest Regression (MAE of 1.23 years for females and 1.65 years for males) and Support Vector Regression with Radial Basis Function Kernel (MAE of 1.47 years for females and 1.72 years for males) which had acceptable and comparable computational efficiency. Findings of the present study could be used as a guide for optimizing methodology when quantifying age-related changes during development.

Download Full-text

Deep Learning versus Conventional Machine Learning for Detection of Healthcare-Associated Infections in French Clinical Narratives

Methods of Information in Medicine ◽

10.1055/s-0039-1677692 ◽

2019 ◽

Vol 58 (01) ◽

pp. 031-041 ◽

Cited By ~ 4

Author(s):

Sara Rabhi ◽

Jérémie Jakubowicz ◽

Marie-Helene Metzger

Keyword(s):

Machine Learning ◽

Deep Learning ◽

Characteristic Curve ◽

Bayesian Optimization ◽

Gradient Boosting ◽

Support Vector ◽

University Hospitals ◽

Infection Control Practitioner ◽

Medical Reports ◽

Conventional Machine

Objective The objective of this article was to compare the performances of health care-associated infection (HAI) detection between deep learning and conventional machine learning (ML) methods in French medical reports. Methods The corpus consisted in different types of medical reports (discharge summaries, surgery reports, consultation reports, etc.). A total of 1,531 medical text documents were extracted and deidentified in three French university hospitals. Each of them was labeled as presence (1) or absence (0) of HAI. We started by normalizing the records using a list of preprocessing techniques. We calculated an overall performance metric, the F1 Score, to compare a deep learning method (convolutional neural network [CNN]) with the most popular conventional ML models (Bernoulli and multi-naïve Bayes, k-nearest neighbors, logistic regression, random forests, extra-trees, gradient boosting, support vector machines). We applied the hyperparameter Bayesian optimization for each model based on its HAI identification performances. We included the set of text representation as an additional hyperparameter for each model, using four different text representations (bag of words, term frequency–inverse document frequency, word2vec, and Glove). Results CNN outperforms all other conventional ML algorithms for HAI classification. The best F1 Score of 97.7% ± 3.6% and best area under the curve score of 99.8% ± 0.41% were achieved when CNN was directly applied to the processed clinical notes without a pretrained word2vec embedding. Through receiver operating characteristic curve analysis, we could achieve a good balance between false notifications (with a specificity equal to 0.937) and system detection capability (with a sensitivity equal to 0.962) using the Youden's index reference. Conclusions The main drawback of CNNs is their opacity. To address this issue, we investigated CNN inner layers' activation values to visualize the most meaningful phrases in a document. This method could be used to build a phrase-based medical assistant algorithm to help the infection control practitioner to select relevant medical records. Our study demonstrated that deep learning approach outperforms other classification learning algorithms for automatically identifying HAIs in medical reports.

Download Full-text

Comparative analysis of a deep learning approach with various classification techniques for credit score computation

Recent Advances in Computer Science and Communications ◽

10.2174/2666255813999200721004720 ◽

2020 ◽

Vol 13 ◽

Author(s):

Arvind Pandey ◽

Shipra Shukla ◽

Krishna Kumar Mohbey

Keyword(s):

Machine Learning ◽

Deep Learning ◽

Comparative Analysis ◽

Predictive Accuracy ◽

Credit Scoring ◽

Gradient Boosting ◽

Learning Approach ◽

Probit Regression ◽

Credit Score ◽

Machine Learning Approach

Background: Large financial companies are perpetually creating and updating customer scoring techniques. From a risk management view, this research for the predictive accuracy of probability is of vital importance than the traditional binary result of classification, i.e., non-credible and credible customers. The customer's default payment in Taiwan is explored for the case study. Objective: The aim is to audit the comparison between the predictive accuracy of the probability of default with various techniques of statistics and machine learning. Method: In this paper, nine predictive models are compared from which the results of the six models are taken into consideration. Deep learning-based H2O, XGBoost, logistic regression, gradient boosting, naïve Bayes, logit model, and probit regression comparative analysis is performed. The software tools such as R and SAS (university edition) is employed for machine learning and statistical model evaluation. Results: Through the experimental study, we demonstrate that XGBoost performs better than other AI and ML algorithms. Conclusion: Machine learning approach such as XGBoost effectively used for credit scoring, among other data mining and statistical approaches.

Download Full-text

Region-specific Brain Age Prediction Models for Children and Adolescents Derived by Machine Learning

10.21203/rs.3.rs-839010/v1 ◽

2021 ◽

Author(s):

Hyerin Jeong ◽

Ukeob Park ◽

Seung Wan Kang

Keyword(s):

Machine Learning ◽

Children And Adolescents ◽

Developmental Disorders ◽

Prediction Models ◽

Brain Regions ◽

Normal Children ◽

Left Posterior ◽

Minimum Number ◽

Brain Age ◽

Selection Of

Abstract EEG biomarkers can reveal significant and actionable differences in brain development between normal children and those with developmental disorders. Frontal slow frequency EEG is one common differentiator between normal and abnormal brain function. The present study sought to establish models, based on machine learning, to predict brain age in children and adolescents. Four brain regions were studied: left anterior, right anterior, left posterior, and right posterior, based on the different functions characteristic of each region. Importantly, differences were also considered in the construction of the models. All models yielded promising r2 values for the prediction of brain age, with values of 0.80 or higher. Our technique employed a tree-based feature selection algorithm, allowing selection of a minimum number of features while still preserving predictive power. These prediction models can be used to quantify deviations between estimated and biological brain age, and so serve as valuable tools in efforts to assess and intervene early in several profound developmental disorders.

Download Full-text

Morphological Neuroimaging Biomarkers for Tinnitus: Evidence Obtained by Applying Machine Learning

Neural Plasticity ◽

10.1155/2019/1712342 ◽

2019 ◽

Vol 2019 ◽

pp. 1-11

Author(s):

Yawen Liu ◽

Haijun Niu ◽

Jianming Zhu ◽

Pengfei Zhao ◽

Hongxia Yin ◽

...

Keyword(s):

Machine Learning ◽

Healthy Subjects ◽

Morphological Changes ◽

Gray Matter Volume ◽

Brain Regions ◽

Classification Model ◽

Middle Temporal Gyrus ◽

Support Vector ◽

Svm Classifier ◽

Neuroimaging Biomarkers

According to previous studies, many neuroanatomical alterations have been detected in patients with tinnitus. However, the results of these studies have been inconsistent. The objective of this study was to explore the cortical/subcortical morphological neuroimaging biomarkers that may characterize idiopathic tinnitus using machine learning methods. Forty-six patients with idiopathic tinnitus and fifty-six healthy subjects were included in this study. For each subject, the gray matter volume of 61 brain regions was extracted as an original feature pool. From this feature pool, a hybrid feature selection algorithm combining the F-score and sequential forward floating selection (SFFS) methods was performed to select features. Then, the selected features were used to train a support vector machine (SVM) model. The area under the curve (AUC) and accuracy were used to assess the performance of the classification model. As a result, a combination of 13 cortical/subcortical brain regions was found to have the highest classification accuracy for effectively differentiating patients with tinnitus from healthy subjects. These brain regions include the bilateral hypothalamus, right insula, bilateral superior temporal gyrus, left rostral middle frontal gyrus, bilateral inferior temporal gyrus, right inferior parietal lobule, right transverse temporal gyrus, right middle temporal gyrus, right cingulate gyrus, and left superior frontal gyrus. The accuracy in the training and test datasets was 80.49% and 80.00%, respectively, and the AUC was 0.8586. To the best of our knowledge, this is the first study to elucidate brain morphological changes in patients with tinnitus by applying an SVM classifier. This study provides validated cortical/subcortical morphological neuroimaging biomarkers to differentiate patients with tinnitus from healthy subjects and contributes to the understanding of neuroanatomical alterations in patients with tinnitus.

Download Full-text

A Comparative Analysis of Novel Deep Learning and Ensemble Learning Models to Predict the Allergenicity of Food Proteins

Foods ◽

10.3390/foods10040809 ◽

2021 ◽

Vol 10 (4) ◽

pp. 809

Author(s):

Liyang Wang ◽

Dantong Niu ◽

Xinjie Zhao ◽

Xiaoya Wang ◽

Mengzhen Hao ◽

...

Keyword(s):

Machine Learning ◽

Deep Learning ◽

Ensemble Learning ◽

Food Allergen ◽

Gradient Boosting ◽

Learning Models ◽

Food Proteins ◽

Deep Model ◽

Allergen Identification ◽

Machine Learning Models

Traditional food allergen identification mainly relies on in vivo and in vitro experiments, which often needs a long period and high cost. The artificial intelligence (AI)-driven rapid food allergen identification method has solved the above mentioned some drawbacks and is becoming an efficient auxiliary tool. Aiming to overcome the limitations of lower accuracy of traditional machine learning models in predicting the allergenicity of food proteins, this work proposed to introduce deep learning model—transformer with self-attention mechanism, ensemble learning models (representative as Light Gradient Boosting Machine (LightGBM) eXtreme Gradient Boosting (XGBoost)) to solve the problem. In order to highlight the superiority of the proposed novel method, the study also selected various commonly used machine learning models as the baseline classifiers. The results of 5-fold cross-validation showed that the area under the receiver operating characteristic curve (AUC) of the deep model was the highest (0.9578), which was better than the ensemble learning and baseline algorithms. But the deep model need to be pre-trained, and the training time is the longest. By comparing the characteristics of the transformer model and boosting models, it can be analyzed that, each model has its own advantage, which provides novel clues and inspiration for the rapid prediction of food allergens in the future.

Download Full-text

Local Epigenomic Data are more Informative than Local Genome Sequence Data in Predicting Enhancer-Promoter Interactions Using Neural Networks

Genes ◽

10.3390/genes11010041 ◽

2019 ◽

Vol 11 (1) ◽

pp. 41 ◽

Cited By ~ 1

Author(s):

Mengli Xiao ◽

Zhong Zhuang ◽

Wei Pan

Keyword(s):

Neural Network ◽

Machine Learning ◽

Deep Learning ◽

Test Data ◽

Sequence Data ◽

Gradient Boosting ◽

Genome Wide Association Studies ◽

Learning Methods ◽

Machine Learning Methods ◽

Local Sequence

Enhancer-promoter interactions (EPIs) are crucial for transcriptional regulation. Mapping such interactions proves useful for understanding disease regulations and discovering risk genes in genome-wide association studies. Some previous studies showed that machine learning methods, as computational alternatives to costly experimental approaches, performed well in predicting EPIs from local sequence and/or local epigenomic data. In particular, deep learning methods were demonstrated to outperform traditional machine learning methods, and using DNA sequence data alone could perform either better than or almost as well as only utilizing epigenomic data. However, most, if not all, of these previous studies were based on randomly splitting enhancer-promoter pairs as training, tuning, and test data, which has recently been pointed out to be problematic; due to multiple and duplicating/overlapping enhancers (and promoters) in enhancer-promoter pairs in EPI data, such random splitting does not lead to independent training, tuning, and test data, thus resulting in model over-fitting and over-estimating predictive performance. Here, after correcting this design issue, we extensively studied the performance of various deep learning models with local sequence and epigenomic data around enhancer-promoter pairs. Our results confirmed much lower performance using either sequence or epigenomic data alone, or both, than reported previously. We also demonstrated that local epigenomic features were more informative than local sequence data. Our results were based on an extensive exploration of many convolutional neural network (CNN) and feed-forward neural network (FNN) structures, and of gradient boosting as a representative of traditional machine learning.

Download Full-text

Image and structured data analysis for prognostication of health outcomes in patients presenting to the Emergency Department during the COVID-19 pandemic

10.1101/2021.07.07.21260097 ◽

2021 ◽

Author(s):

Liam Butler ◽

Ibrahim Karabayir ◽

Mohammad Samie Tootooni ◽

Majid Afshar ◽

Ari Goldberg ◽

...

Keyword(s):

Machine Learning ◽

Emergency Department ◽

Intensive Care ◽

Deep Learning ◽

Clinical Data ◽

Chest Radiographs ◽

Gradient Boosting ◽

Learning Models ◽

Icu Admission ◽

Light Gradient

Background: Patients admitted to the emergency department (ED) with COVID-19 symptoms are routinely required to have chest radiographs and computed tomography (CT) scans. COVID-19 infection has been directly related to development of acute respiratory distress syndrome (ARDS) and severe infections lead to admission to intensive care and can also lead to death. The use of clinical data in machine learning models available at time of admission to ED can be used to assess possible risk of ARDS, need for intensive care unit (ICU) admission as well as risk of mortality. In addition, chest radiographs can be inputted into a deep learning model to further assess these risks. Purpose: This research aimed to develop machine and deep learning models using both structured clinical data and image data from the electronic health record (EHR) to adverse outcomes following ED admission. Materials and Methods: Light Gradient Boosting Machines (LightGBM) was used as the main machine learning algorithm using all clinical data including 42 variables. Compact models were also developed using 15 the most important variables to increase applicability of the models in clinical settings. To predict risk of the aforementioned health outcome events, transfer learning from the CheXNet model was implemented on our data as well. This research utilized clinical data and chest radiographs of 3571 patients 18 years and older admitted to the emergency department between 9th March 2020 and 29th October 2020 at Loyola University Medical Center. Main Findings: Our research results show that we can detect COVID-19 infection (AUC = 0.790 (0.746-0.835)) and predict the risk of developing ARDS (AUC = 0.781 (0.690-0.872), ICU admission (AUC = 0.675 (0.620-0.713)), and mortality (AUC = 0.759 (0.678-0.840)) at moderate accuracy from both chest X-ray images and clinical data. Principal Conclusions: The results can help in clinical decision making, especially when addressing ARDS and mortality, during the assessment of patients admitted to the ED with or without COVID-19 symptoms.

Download Full-text

Recent Development of Artificial Intelligence Applied in Echocardiography: A Review (Preprint)

10.2196/preprints.17227 ◽

2019 ◽

Author(s):

Lu Liu ◽

Ahmed Elazab ◽

Baiying Lei ◽

Tianfu Wang

Keyword(s):

Artificial Intelligence ◽

Machine Learning ◽

Deep Learning ◽

Cardiovascular Diseases ◽

Disease Diagnosis ◽

Machine Learning Techniques ◽

Tissue Segmentation ◽

The Past ◽

Learning Techniques ◽

The Future

BACKGROUND Echocardiography has a pivotal role in the diagnosis and management of cardiovascular diseases since it is real-time, cost-effective, and non-invasive. The development of artificial intelligence (AI) techniques have led to more intelligent and automatic computer-aided diagnosis (CAD) systems in echocardiography over the past few years. Automatic CAD mainly includes classification, detection of anatomical structures, tissue segmentation, and disease diagnosis, which are mainly completed by machine learning techniques and the recent developed deep learning techniques. OBJECTIVE This review aims to provide a guide for researchers and clinicians on relevant aspects of AI, machine learning, and deep learning. In addition, we review the recent applications of these methods in echocardiography and identify how echocardiography could incorporate AI in the future. METHODS This paper first summarizes the overview of machine learning and deep learning. Second, it reviews current use of AI in echocardiography by searching literature in the main databases for the past 10 years and finally discusses potential limitations and challenges in the future. RESULTS AI has showed promising improvements in analysis and interpretation of echocardiography to a new stage in the fields of standard views detection, automated analysis of chamber size and function, and assessment of cardiovascular diseases. CONCLUSIONS Compared with machine learning, deep learning methods have achieved state-of-the-art performance across different applications in echocardiography. Although there are challenges such as the required large dataset, AI can provide satisfactory results by devising various strategies. We believe AI has the potential to improve accuracy of diagnosis, reduce time consumption, and decrease the load of cardiologists.

Download Full-text

Grey matter biomarker identification in Schizophrenia: detecting regional alterations and their underlying substrates

10.1101/357954 ◽

2018 ◽

Author(s):

V. Chatzi ◽

R.P. Teixeira ◽

J. Shawe-Taylor ◽

A. Altmann ◽

O. O’Daly ◽

...

Keyword(s):

Machine Learning ◽

Grey Matter ◽

Structural Mri ◽

Brain Regions ◽

List Type ◽

Healthy Controls ◽

Learning Approaches ◽

Learning Framework ◽

Software Packages ◽

Biological Substrates

AbstractState-of-the-art approaches in Schizophrenia research investigate neuroanatomical biomarkers using structural Magnetic Resonance Imaging. However, current models are 1) voxel-wise, 2) difficult to interpret in biologically meaningful ways, and 3) difficult to replicate across studies. Here, we propose a machine learning framework that enables the identification of sparse, region-wise grey matter neuroanatomical biomarkers and their underlying biological substrates by integrating well-established statistical and machine learning approaches. We address the computational issues associated with application of machine learning on structural MRI data in Schizophrenia, as discussed in recent reviews, while promoting transparent science using widely available data and software. In this work, a cohort of patients with Schizophrenia and healthy controls was used. It was found that the cortical thickness in left pars orbitalis seems to be the most reliable measure for distinguishing patients with Schizophrenia from healthy controls.HighlightsWe present a sparse machine learning framework to identify biologically meaningful neuroanatomical biomarkers for SchizophreniaOur framework addresses methodological pitfalls associated with application of machine learning on structural MRI data in Schizophrenia raised by several recent reviewsOur pipeline is easy to replicate using widely available software packagesThe presented framework is geared towards identification of specific changes in brain regions that relate directly to the pathology rather than classification per se

Download Full-text