Multi-class AUC Research Articles

Ensemble of machine learning techniques to predict survival in kidney transplant recipients

Kidney transplant recipients face a high cardiovascular risk, which is a leading cause of death in this patient group. This article proposes the application of clustering techniques and feature selection to predict the survival outcomes of kidney transplant recipients based on machine learning techniques and mainstream statistical methods. First, feature selection techniques (Boruta, Random Survival Forest and Elastic Net) are used to detect the most relevant variables. Subsequently, each set of variables obtained by each feature selection technique is used as input for the clustering algorithms used (Consensus Clustering, Self-Organizing Map and Agglomerative Clustering) to determine which combination of feature selection, clustering algorithm and number of clusters maximizes intercluster variability. Next, the mechanism called False Clustering Discovery Reduction is applied to obtain the minimum number of statistically differentiable populations after applying a control metric. This metric is based on a variance test to confirm that reducing the number of clusters does not generate significant losses in the heterogeneity obtained. This approach was applied to the Organ Procurement and Transplantation Network medical dataset (n = 11,332). The combination of Random Survival Forest and consensus clustering yielded the optimal result of 4 clusters starting from 8 initial ones. Finally, for each population, Kaplan-Meier survival curves are generated to predict the survival of new patients based on the predictions of the XGBoost classifier, with an overall multi-class AUC of 98.11%.

Multi-class AUC maximization for imbalanced ordinal multi-stage tropical cyclone intensity change forecast

Intense tropical cyclones (TCs) cause significant damage to human societies. Forecasting the multiple stages of TC intensity changes is considerably crucial yet challenging. This difficulty arises due to imbalanced data distribution and the need for ordinal multi-class classification. While existing classification methods, such as linear discriminant analysis, have been utilized to predict rare rapidly intensifying (RI) stages based on features related TC intensity changes, they are limited to binary classification distinguishing between RI and non-RI stages. In this paper, we introduce a novel methodology to tackle the challenges of imbalanced ordinal multi-class classification. We extend the Area Under the Curve maximization technique with inter-instance/class cross-hinge losses and inter-class distance-based slack variables. The proposed loss function, implemented within a deep learning framework, demonstrates its effectiveness using real sequence data of multi-stage TC intensity changes, including satellite infrared images and environmental variables observed in the western North Pacific.

Comparing multi-class classifier performance by multi-class ROC analysis: A nonparametric approach

The area under the Receiver Operating Characteristic (ROC) curve (AUC) is a standard metric for quantifying and comparing binary classifiers. Real world applications often require classification into multiple (more than two) classes. For multi-class classifiers that produce class membership scores, a popular multi-class AUC (MAUC) variant is to average the pairwise AUC values (Hand and Till, 2001). Due to the complicated correlation patterns, the variance of MAUC is often estimated numerically using resampling techniques. This work is a generalization of DeLong’s nonparameteric approach for binary AUC analysis (DeLong et al., 1988) to MAUC. We first derive the closed-form expression of the covariance matrix of the pairwise AUCs within a single MAUC. Then by dropping higher order terms, we obtain an approximate covariance matrix with a compact, matrix factorization form, which then serves as the basis for variance estimation of a single MAUC. We further extend this approach to estimate the covariance of correlated MAUCs that arise from multiple competing classifiers. For the special case of binary correlated AUCs, our results coincide with that of DeLong. Our numerical studies confirm the accuracy of the variance and covariance estimates. We provide the source code of the proposed covariance estimation of correlated MAUCs on GitHub (https://tinyurl.com/euj6wvsz) for its easy adoption by machine learning and statistical analysis packages to quantify and compare multi-class classifiers.

AI-based classification of three common malignant tumors in neuro-oncology: A multi-institutional comparison of machine learning and deep learning methods

PurposeTo determine if machine learning (ML) or deep learning (DL) pipelines perform better in AI-based three-class classification of glioblastoma (GBM), intracranial metastatic disease (IMD) and primary CNS lymphoma (PCNSL). MethodologyRetrospective analysis included 502 cases for training (208 GBM, 67 PCNSL and 227 IMD), with external validation on 86 cases (27:27:32). Multiparametric MRI images (T1W, T2W, FLAIR, DWI and T1-CE) were co-registered, resampled, denoised and intensity normalized, followed by semiautomatic 3D segmentation of the enhancing tumor (ET) and peritumoral region (PTR). Model performance was assessed using several ML pipelines and 3D-convolutional neural networks (3D-CNN) using sequence specific masks, as well as combination of masks. All pipelines were trained and evaluated with 5-fold nested cross-validation on internal data followed by external validation using multi-class AUC. ResultsTwo ML models achieved similar performance on test set, one using T2-ET and T2-PTR masks (AUC: 0.885, 95% CI: [0.816, 0.935] and another using T1-CE-ET and FLAIR-PTR mask (AUC: 0.878, CI: [0.804, 0.930]). The best performing DL models achieved an AUC of 0.854, (CI [0.774, 0.914]) on external data using T1-CE-ET and T2-PTR masks, followed by model derived from T1-CE-ET, ADC-ET and FLAIR-PTR masks (AUC: 0.851, CI [0.772, 0.909]). ConclusionBoth ML and DL derived pipelines achieved similar performance. T1-CE mask was used in three of the top four overall models. Additionally, all four models had some mask derived from PTR, either T2WI or FLAIR.

Towards Decision-Friendly AUC: Learning Multi-Classifier with AUCµ

Area Under the ROC Curve (AUC) is a widely used ranking metric in imbalanced learning due to its insensitivity to label distributions. As a well-known multiclass extension of AUC, Multiclass AUC (MAUC, a.k.a. M-metric) measures the average AUC of multiple binary classifiers. In this paper, we argue that simply optimizing MAUC is far from enough for imbalanced multi-classification. More precisely, MAUC only focuses on learning scoring functions via ranking optimization, while leaving the decision process unconsidered. Therefore, scoring functions being able to make good decisions might suffer from low performance in terms of MAUC. To overcome this issue, we turn to explore AUCµ, another multiclass variant of AUC, which further takes the decision process into consideration. Motivated by this fact, we propose a surrogate risk optimization framework to improve model performance from the perspective of AUCµ. Practically, we propose a two-stage training framework for multi-classification, where at the first stage a scoring function is learned maximizing AUCµ, and at the second stage we seek for a decision function to improve the F1-metric via our proposed soft F1. Theoretically, we first provide sufficient conditions that optimizing the surrogate losses could lead to the Bayes optimal scoring function. Afterward, we show that the proposed surrogate risk enjoys a generalization bound in order of O(1/√N). Experimental results on four benchmark datasets demonstrate the effectiveness of our proposed method in both AUCµ and F1-metric.

Assessing the surveillance use of 2018 EFP/AAP classification of periodontitis: A validation study and clustering analysis.

The performance of the 2018 European Federation of Periodontology/American Academy of Periodontology (EFP/AAP) classification of periodontitis for epidemiology surveillance purposes remains to be investigated. This study assessed the surveillance use of the 2018 EFP/AAP classification and its agreement with the unsupervised clustering method compared with the 2012 Centers for Disease Control and Prevention(CDC)/AAP case definition. Participants (n=9424) in the National Health and Nutrition Examination Survey (NHANES) were staged by the 2018 EFP/AAP classification and classified into subgroups via k-medoids clustering. Concordance levels between periodontitis definitions and the clustering method were evaluated through the multiclass area under the receiver operating characteristic curve (multiclass AUC) among "periodontitis cases" and the general population, respectively. The multiclass AUC of the 2012 CDC/AAP definition versus clustering was used as a reference. The associations of periodontitis with chronic diseases were estimated using multivariable logistic regression. All the participants were identified as "periodontitis cases" by the 2018 EFP/AAP classification, and the prevalence of stage III-IV was 30%. The optimal numbers of clusters were three and four. The 2012 CDC/AAP definition versus clustering yielded a multiclass AUC of 0.82 and 0.85 among the general population and "periodontitis cases," respectively. The multiclass AUC of the 2018 EFP/AAP classification versus clustering was 0.77 and 0.78 for different target populations. Similar patterns prevailed in associations with chronic diseases between the 2018 EFP/AAP classification and clustering. The validity of the 2018 EFP/AAP classification was verified by the unsupervised clustering method, which performed better in distinguishing "periodontitis cases" than classifying the general population. For surveillance purposes, the 2012 CDC/AAP definition showed a higher agreement level with the clustering method than the 2018 EFP/AAP classification.

Identification and validation of IgG N-glycosylation biomarkers of esophageal carcinoma.

Altered Immunoglobulin G (IgG) N-glycosylation is associated with aging, inflammation, and diseases status, while its effect on esophageal squamous cell carcinoma (ESCC) remains unknown. As far as we know, this is the first study to explore and validate the association of IgG N-glycosylation and the carcinogenesis progression of ESCC, providing innovative biomarkers for the predictive identification and targeted prevention of ESCC. In total, 496 individuals of ESCC (n=114), precancerosis (n=187) and controls (n=195) from the discovery population (n=348) and validation population (n=148) were recruited in the study. IgG N-glycosylation profile was analyzed and an ESCC-related glycan score was composed by a stepwise ordinal logistic model in the discovery population. The receiver operating characteristic (ROC) curve with the bootstrapping procedure was used to assess the performance of the glycan score. In the discovery population, the adjusted OR of GP20 (digalactosylated monosialylated biantennary with core and antennary fucose), IGP33 (the ratio of all fucosylated monosyalilated and disialylated structures), IGP44 (the proportion of high mannose glycan structures in total neutral IgG glycans), IGP58 (the percentage of all fucosylated structures in total neutral IgG glycans), IGP75 (the incidence of bisecting GlcNAc in all fucosylated digalactosylated structures in total neutral IgG glycans), and the glycan score are 4.03 (95% CI: 3.03-5.36, P<0.001), 0.69 (95% CI: 0.55-0.87, P<0.001), 0.56 (95% CI: 0.45-0.69, P<0.001), 0.52 (95% CI: 0.41-0.65, P<0.001), 7.17 (95% CI: 4.77-10.79, P<0.001), and 2.86 (95% CI: 2.33-3.53, P<0.001), respectively. Individuals in the highest tertile of the glycan score own an increased risk (OR: 11.41), compared with those in the lowest. The average multi-class AUC are 0.822 (95% CI: 0.786-0.849). Findings are verified in the validation population, with an average AUC of 0.807 (95% CI: 0.758-0.864). Our study demonstrated that IgG N-glycans and the proposed glycan score appear to be promising predictive markers for ESCC, contributing to the early prevention of esophageal cancer. From the perspective of biological mechanism, IgG fucosylation and mannosylation might involve in the carcinogenesis progression of ESCC, and provide potential therapeutic targets for personalized interventions of cancer progression.

MRI-based basal forebrain atrophy and volumetric signatures associated with limbic TDP-43 compared to Alzheimer's disease pathology

BackgroundIt is not clear to which degree limbic TDP-43 pathology associates with a cholinergic deficit in the absence of Alzheimer's disease (AD) pathology. ObjectiveReplicate and extend recent evidence on cholinergic basal forebrain atrophy in limbic TDP-43 and evaluate MRI based patterns of atrophy as a surrogate marker for TDP-43. MethodsWe studied ante-mortem MRI data of 11 autopsy cases with limbic TDP-43 pathology, 47 cases with AD pathology, and 26 mixed AD/TDP-43 cases from the ADNI autopsy sample, and 17 TDP-43, 170 AD, and 58 mixed AD/TDP-43 cases from the NACC autopsy sample. Group differences in basal forebrain and other brain volumes of interest were assessed using Bayesian ANCOVA. We assessed the diagnostic utility of MRI based patterns of brain atrophy using voxel-based receiver operating characteristics and random forest analyses. ResultsIn the NACC sample, we found moderate evidence for the absence of a difference in basal forebrain volumes between AD, TDP-43, and mixed pathologies (Bayes factor(BF)10 = 0.324), and very strong evidence for lower hippocampus volume in TDP-43 and mixed cases compared with AD cases (BF10 = 156.1). The ratio of temporal to hippocampus volume reached an AUC of 75% for separating pure TDP-43 from pure AD cases. Random-forest analysis between TDP-43, AD, and mixed pathology reached only a multiclass AUC of 0.63 based on hippocampus, middle-inferior temporal gyrus, and amygdala volumes. Findings in the ADNI sample were consistent with these results. ConclusionA comparable degree of basal forebrain atrophy in pure TDP-43 cases compared to AD cases encourages studies on the effect of cholinergic treatment in amnestic dementia due to TDP-43. A distinct pattern of temporo-limbic brain atrophy may serve as a surrogate marker to enrich samples in clinical trials for the presence of TDP-43 pathology.

HiPerMAb: a tool for judging the potential of small sample size biomarker pilot studies.

Common statistical approaches are not designed to deal with so-called "short fat data" in biomarker pilot studies, where the number of biomarker candidates exceeds the sample size by magnitudes. High-throughput technologies for omics data enable the measurement of ten thousands and more biomarker candidates for specific diseases or states of a disease. Due to the limited availability of study participants, ethical reasons and high costs for sample processing and analysis researchers often prefer to start with a small sample size pilot study in order to judge the potential of finding biomarkers that enable - usually in combination - a sufficiently reliable classification of the disease state under consideration. We developed a user-friendly tool, called HiPerMAb that allows to evaluate pilot studies based on performance measures like multiclass AUC, entropy, area above the cost curve, hypervolume under manifold, and misclassification rate using Monte-Carlo simulations to compute the p-values and confidence intervals. The number of "good" biomarker candidates is compared to the expected number of "good" biomarker candidates in a data set with no association to the considered disease states. This allows judging the potential in the pilot study even if statistical tests with correction for multiple testing fail to provide any hint of significance.

A new approach to predicting mortality in dialysis patients using sociodemographic features based on artificial intelligence

One of the main problems that affect patients in dialysis therapy who are on the waiting list to receive a kidney transplant is predicting their survival time if they do not receive a transplant. This paper proposes a new approach to survival prediction based on artificial intelligence techniques combined with statistical methods to study the association between sociodemographic factors and patient survival on the waiting list if they do not receive a kidney transplant. This new approach consists of a first stage that uses the clustering techniques that are best suited to the data structure (K-Means, Mini Batch K-Means, Agglomerative Clustering and K-Modes) used to identify the risk profile of dialysis patients. Later, a new method called False Clustering Discovery Reduction is performed to determine the minimum number of populations to be studied, and whose mortality risk is statistically differentiable. This approach was applied to the OPTN medical dataset (n = 44,663). The procedure started from 11 initial clusters obtained with the Agglomerative technique, and was reduced to eight final risk populations, for which their Kaplan-Meier survival curves were provided. With this result, it is possible to make predictions regarding the survival time of a new patient who enters the waiting list if the sociodemographic profile of the patient is known. To do so, the predictive algorithm XGBoost is used, which allows the cluster to which it belongs to be predicted and the corresponding Kaplan-Meier curve to be associated with it. This prediction process is achieved with an overall Multi-class AUC of 99.08 %.

Learning With Multiclass AUC: Theory and Algorithms.

The Area under the ROC curve (AUC) is a well-known ranking metric for problems such as imbalanced learning and recommender systems. The vast majority of existing AUC-optimization-based machine learning methods only focus on binary-class cases, while leaving the multiclass cases unconsidered. In this paper, we start an early trial to consider the problem of learning multiclass scoring functions via optimizing multiclass AUC metrics. Our foundation is based on the M metric, which is a well-known multiclass extension of AUC. We first pay a revisit to this metric, showing that it could eliminate the imbalance issue from the minority class pairs. Motivated by this, we propose an empirical surrogate risk minimization framework to approximately optimize the M metric. Theoretically, we show that: (i) optimizing most of the popular differentiable surrogate losses suffices to reach the Bayes optimal scoring function asymptotically; (ii) the training framework enjoys an imbalance-aware generalization error bound, which pays more attention to the bottleneck samples of minority classes compared with the traditional O(√{1/N}) result. Practically, to deal with the low scalability of the computational operations, we propose acceleration methods for three popular surrogate loss functions, including the exponential loss, squared loss, and hinge loss, to speed up loss and gradient evaluations. Finally, experimental results on 11 real-world datasets demonstrate the effectiveness of our proposed framework. The code is now available at https://github.com/joshuaas/Learning-with-Multiclass-AUC-Theory-and-Algorithms.

Endoscopic Surgical Treatment of Osteoarthritis and Prognostic Model Construction.

Purpose Osteoarthritis (OA) is a degenerative disease of joints. Currently, there is still a lack of effective tools to predict the long-term efficacy of surgical treatment of OA. The purpose of this study was to explore the prognostic factors of endoscopic surgery for OA and to predict the long-term efficacy of this type of surgery for OA by establishing a prognostic model. Methods Baseline and follow-up data on 236 OA patients who underwent surgery in our hospital from January 2017 to December 2021 were selected and patients were randomly assigned to a training set (n = 165) and a test set (n = 71). The Pearson correlation coefficient was used to analyze the correlation between features. Feature selection was performed by recursive feature elimination (RFE) and linear regression. K-means clustering analysis was performed on the selected features to obtain the number of output layers. Finally, a single hidden layer error backpropagation (BP)-artificial neural network (ANN) model was established on the training set, and receiver operating characteristic (ROC) curve was drawn on the test set for verification. Results Correlation analysis revealed no redundancy among features. RFE and linear regression screened out the features associated with postoperative prognosis under endoscopic surgery: sex, age, BMI, region, morning stiffness time, step count, and osteophyte area. K-means clustering yielded that the optimal number of categories was three, the same as the number of categories for the outcome variable. Therefore, a 7-1-3 BP neural network model was established based on these 7 features, and this model could predict the postoperative situation within one year to a relatively accurate extent: area under curve values (AUC) were 0.814, 0.700, and 0.761 in patients with worse, unchanged, and improved conditions one year after surgery, respectively, higher than the multiclass AUC value (0.646). Conclusion The prognostic model of endoscopic surgery for OA constructed in this study can well predict the disease progression of patients within one year after surgery.

18F]FDG-PET/CT radiomics for the identification of genetic clusters in pheochromocytomas and paragangliomas

ObjectivesBased on germline and somatic mutation profiles, pheochromocytomas and paragangliomas (PPGLs) can be classified into different clusters. We investigated the use of [18F]FDG-PET/CT radiomics, SUVmax and biochemical profile for the identification of the genetic clusters of PPGLs.MethodsIn this single-centre cohort, 40 PPGLs (13 cluster 1, 18 cluster 2, 9 sporadic) were delineated using a 41% adaptive threshold of SUVpeak ([18F]FDG-PET) and manually (low-dose CT; ldCT). Using PyRadiomics, 211 radiomic features were extracted. Stratified 5-fold cross-validation for the identification of the genetic cluster was performed using multinomial logistic regression with dimensionality reduction incorporated per fold. Classification performances of biochemistry, SUVmax and PET(/CT) radiomic models were compared and presented as mean (multiclass) test AUCs over the five folds. Results were validated using a sham experiment, randomly shuffling the outcome labels.ResultsThe model with biochemistry only could identify the genetic cluster (multiclass AUC 0.60). The three-factor PET model had the best classification performance (multiclass AUC 0.88). A simplified model with only SUVmax performed almost similarly. Addition of ldCT features and biochemistry decreased the classification performances. All sham AUCs were approximately 0.50.ConclusionPET radiomics achieves a better identification of PPGLs compared to biochemistry, SUVmax, ldCT radiomics and combined approaches, especially for the differentiation of sporadic PPGLs. Nevertheless, a model with SUVmax alone might be preferred clinically, weighing model performances against laborious radiomic analysis. The limited added value of radiomics to the overall classification performance for PPGL should be validated in a larger external cohort.Key Points• Radiomics derived from [18F]FDG-PET/CT has the potential to improve the identification of the genetic clusters of pheochromocytomas and paragangliomas.• A simplified model with SUVmaxonly might be preferred clinically, weighing model performances against the laborious radiomic analysis.• Cluster 1 and 2 PPGLs generally present distinctive characteristics that can be captured using [18F]FDG-PET imaging. Sporadic PPGLs appear more heterogeneous, frequently resembling cluster 2 PPGLs and occasionally resembling cluster 1 PPGLs.

Triple-Negative Breast Cancer Analysis Based on Metabolic Gene Classification and Immunotherapy.

Triple negative breast cancer (TNBC) has negative expression of ER, PR and HER-2. TNBC shows high histological grade and positive rate of lymph node metastasis, easy recurrence and distant metastasis. Molecular typing based on metabolic genes can reflect deeper characteristics of breast cancer and provide support for prognostic evaluation and individualized treatment. Metabolic subtypes of TNBC samples based on metabolic genes were determined by consensus clustering. CIBERSORT method was applied to evaluate the score distribution and differential expression of 22 immune cells in the TNBC samples. Linear discriminant analysis (LDA) established a subtype classification feature index. Kaplan-Meier (KM) and receiver operating characteristic (ROC) curves were generated to validate the performance of prognostic metabolic subtypes in different datasets. Finally, we used weighted correlation network analysis (WGCNA) to cluster the TCGA expression profile dataset and screen the co-expression modules of metabolic genes. Consensus clustering of the TCGA cohort/dataset obtained three metabolic subtypes (MC1, MC2, and MC3). The ROC analysis showed a high prognostic performance of the three clusters in different datasets. Specifically, MC1 had the optimal prognosis, MC3 had a poor prognosis, and the three metabolic subtypes had different prognosis. Consistently, the immune characteristic index established based on metabolic subtypes demonstrated that compared with the other two subtypes, MC1 had a higher IFNγ score, T cell lytic activity and lower angiogenesis score, T cell dysfunction and rejection score. TIDE analysis showed that MC1 patients were more likely to benefit from immunotherapy. MC1 patients were more sensitive to immune checkpoint inhibitors and traditional chemotherapy drugs Cisplatin, Paclitaxel, Embelin, and Sorafenib. Multiclass AUC based on RNASeq and GSE datasets were 0.85 and 0.85, respectively. Finally, based on co-expression network analysis, we screened 7 potential gene markers related to metabolic characteristic index, of which CLCA2, REEP6, SPDEF, and CRAT can be used to indicate breast cancer prognosis. Molecular classification related to TNBC metabolism was of great significance for comprehensive understanding of the molecular pathological characteristics of TNBC, contributing to the exploration of reliable markers for early diagnosis of TNBC and predicting metastasis and recurrence, improvement of the TNBC staging system, guiding individualized treatment.

Quantitative CT and machine learning classification of fibrotic interstitial lung diseases.

To evaluate quantitative computed tomography (QCT) features and QCT feature-based machine learning (ML) models in classifying interstitial lung diseases (ILDs). To compare QCT-ML and deep learning (DL) models' performance. We retrospectively identified 1085 patients with pathologically proven usual interstitial pneumonitis (UIP), nonspecific interstitial pneumonitis (NSIP), and chronic hypersensitivity pneumonitis (CHP) who underwent peri-biopsy chest CT. Kruskal-Wallis test evaluated QCT feature associations with each ILD. QCT features, patient demographics, and pulmonary function test (PFT) results trained eXtreme Gradient Boosting (training/validation set n = 911) yielding 3 models: M1 = QCT features only; M2 = M1 plus age and sex; M3 = M2 plus PFT results. A DL model was also developed. ML and DL model areas under the receiver operating characteristic curve (AUC) and 95% confidence intervals (CIs) were compared for multiclass (UIP vs. NSIP vs. CHP) and binary (UIP vs. non-UIP) classification performances. The majority (69/78 [88%]) of QCT features successfully differentiated the 3 ILDs (adjusted p ≤ 0.05). All QCT-ML models achieved higher AUC than the DL model (multiclass AUC micro-averages 0.910, 0.910, 0.925, and 0.798 and macro-averages 0.895, 0.893, 0.925, and 0.779 for M1, M2, M3, and DL respectively; binary AUC 0.880, 0.899, 0.898, and 0.869 for M1, M2, M3, and DL respectively). M3 demonstrated statistically significant better performance compared to M2 (∆AUC: 0.015, CI: [0.002, 0.029]) for multiclass prediction. QCT features successfully differentiated pathologically proven UIP, NSIP, and CHP. While QCT-based ML models outperformed a DL model for classifying ILDs, further investigations are warranted to determine if QCT-ML, DL, or a combination will be superior in ILD classification. • Quantitative CT features successfully differentiated pathologically proven UIP, NSIP, and CHP. • Our quantitative CT-based machine learning models demonstrated high performance in classifying UIP, NSIP, and CHP histopathology, outperforming a deep learning model. • While our quantitative CT-based machine learning models performed better than a DL model, additional investigations are needed to determine whether either or a combination of both approaches delivers superior diagnostic performance.

Differentiation of hepatocellular adenoma by subtype and hepatocellular carcinoma in non-cirrhotic liver by fractal analysis of perfusion MRI

BackgroundTo investigate whether fractal analysis of perfusion differentiates hepatocellular adenoma (HCA) subtypes and hepatocellular carcinoma (HCC) in non-cirrhotic liver by quantifying perfusion chaos using four-dimensional dynamic contrast-enhanced magnetic resonance imaging (4D-DCE-MRI).ResultsA retrospective population of 63 patients (47 female) with histopathologically characterized HCA and HCC in non-cirrhotic livers was investigated. Our population consisted of 13 hepatocyte nuclear factor (HNF)-1α-inactivated (H-HCAs), 7 β-catenin-exon-3-mutated (bex3-HCAs), 27 inflammatory HCAs (I-HCAs), and 16 HCCs. Four-dimensional fractal analysis was applied to arterial, portal venous, and delayed phases of 4D-DCE-MRI and was performed in lesions as well as remote liver tissue. Diagnostic accuracy of fractal analysis was compared to qualitative MRI features alone and their combination using multi-class diagnostic accuracy testing including kappa-statistics and area under the receiver operating characteristic curve (AUC). Fractal analysis allowed quantification of perfusion chaos, which was significantly different between lesion subtypes (multi-class AUC = 0.90, p < 0.001), except between I-HCA and HCC. Qualitative MRI features alone did not allow reliable differentiation between HCA subtypes and HCC (κ = 0.35). However, combining qualitative MRI features and fractal analysis reliably predicted the histopathological diagnosis (κ = 0.89) and improved differentiation of high-risk lesions (i.e., HCCs, bex3-HCAs) and low-risk lesions (H-HCAs, I-HCAs) from sensitivity and specificity of 43% (95% confidence interval [CI] 23–66%) and 47% (CI 32–64%) for qualitative MRI features to 96% (CI 78–100%) and 68% (CI 51–81%), respectively, when adding fractal analysis.ConclusionsCombining qualitative MRI features with fractal analysis allows identification of HCA subtypes and HCCs in patients with non-cirrhotic livers and improves differentiation of lesions with high and low risk for malignant transformation.

A machine-learning based objective measure for ALS disease severity

Amyotrophic Lateral Sclerosis (ALS) disease severity is usually measured using the subjective, questionnaire-based revised ALS Functional Rating Scale (ALSFRS-R). Objective measures of disease severity would be powerful tools for evaluating real-world drug effectiveness, efficacy in clinical trials, and for identifying participants for cohort studies. We developed a machine learning (ML) based objective measure for ALS disease severity based on voice samples and accelerometer measurements from a four-year longitudinal dataset. 584 people living with ALS consented and carried out prescribed speaking and limb-based tasks. 542 participants contributed 5814 voice recordings, and 350 contributed 13,009 accelerometer samples, while simultaneously measuring ALSFRS-R scores. Using these data, we trained ML models to predict bulbar-related and limb-related ALSFRS-R scores. On the test set (n = 109 participants) the voice models achieved a multiclass AUC of 0.86 (95% CI, 0.85–0.88) on speech ALSFRS-R prediction, whereas the accelerometer models achieved a median multiclass AUC of 0.73 on 6 limb-related functions. The correlations across functions observed in self-reported ALSFRS-R scores were preserved in ML-derived scores. We used these models and self-reported ALSFRS-R scores to evaluate the real-world effects of edaravone, a drug approved for use in ALS. In the cohort of 54 test participants who received edaravone as part of their usual care, the ML-derived scores were consistent with the self-reported ALSFRS-R scores. At the individual level, the continuous ML-derived score can capture gradual changes that are absent in the integer ALSFRS-R scores. This demonstrates the value of these tools for assessing disease severity and, potentially, drug effects.

Predicting non-response to multimodal day clinic treatment in severely impaired depressed patients: a machine learning approach

A considerable number of depressed patients do not respond to treatment. Accurate prediction of non-response to routine clinical care may help in treatment planning and improve results. A longitudinal sample of N = 239 depressed patients was assessed at admission to multi-modal day clinic treatment, after six weeks, and at discharge. First, patient’s treatment response was modelled by identifying longitudinal trajectories using the Hamilton Depression Rating Scale (HDRS-17). Then, individual items of the HDRS-17 at admission as well as individual patient characteristics were entered as predictors of response/non-response trajectories into the binary classification model (eXtremeGradient Boosting; XGBoost). The model was evaluated on a hold-out set and explained in human-interpretable form by SHapley Additive explanation (SHAP) values. The prediction model yielded a multi-class AUC = 0.80 in the hold-out set. The predictive power for the binary classification yielded an AUC = 0.83 (sensitivity = .80, specificity = .77). Most relevant predictors for non-response were insomnia symptoms, younger age, anxiety symptoms, depressed mood, being unemployed, suicidal ideation and somatic symptoms of depressive disorder. Non-responders to routine treatment for depression can be identified and screened for potential next-generation treatments. Such predictors may help personalize treatment and improve treatment response.

Aberrant Structure MRI in Parkinson's Disease and Comorbidity with Depression Based on Multinomial Tensor Regression Analysis.

Background: Depression is a prominent and highly prevalent nonmotor feature in patients with Parkinson’s disease (PD). The neural and pathophysiologic mechanisms of PD with depression (DPD) remain unclear. The current diagnosis of DPD largely depends on clinical evaluation. Methods: We proposed a new family of multinomial tensor regressions that leveraged whole-brain structural magnetic resonance imaging (MRI) data to discriminate among 196 non-depressed PD (NDPD) patients, 84 DPD patients, 200 healthy controls (HC), and to assess the special brain microstructures in NDPD and DPD. The method of maximum likelihood estimation coupled with state-of-art gradient descent algorithms was used to predict the individual diagnosis of PD and the development of DPD in PD patients. Results: The results reveal that the proposed efficient approach not only achieved a high prediction accuracy (0.94) with a multi-class AUC (0.98) for distinguishing between NDPD, DPD, and HC on the testing set but also located the most discriminative regions for NDPD and DPD, including cortical regions, the cerebellum, the brainstem, the bilateral basal ganglia, and the thalamus and limbic regions. Conclusions: The proposed imaging technique based on tensor regression performs well without any prior feature information, facilitates a deeper understanding into the abnormalities in DPD and PD, and plays an essential role in the statistical analysis of high-dimensional complex MRI imaging data to support the radiological diagnosis of comorbidity of depression with PD.

Learning from EMG: semi-automated grading of facial nerve function

The current grading of facial nerve function is based on subjective impression with the established assessment scale of House and Brackmann (HB). Especially for research a more objective method is needed to lower the interobserver variability to a minimum. We developed a semi-automated grading system based on (facial) surface EMG-data measuring the facial nerve function of 28 patients with vestibular schwannoma surgery. The sEMG was recorded preoperatively, postoperatively and after 3–12 months. In addition, the HB grade was determined. After manual selection and preprocessing, the data were subjected to machine learning classificators (Logistic regression, SVM and KNN). Lateralization indices were calculated and multivariant machine learning analysis was performed according to three scenarios [differentiation of normal (1) and slight (2) vs. impaired facial nerve function and classification of HB 1-3 (3)]. The calculated AUC for each scenario showed overall good differentiation capability with a median AUC of 0.72 for scenario 1, 0.91 for scenario 2 and multiclass AUC of 0.74 for scenario 3. This study approach using sEMG and machine learning shows feasibility regarding facial nerve grading in perioperative VS-surgery setting. sEMG may be a viable alternative to House Brackmann regarding objective evaluation of facial function especially for research purposes.