A Comparative Study of the Use of Stratified Cross-Validation and Distribution-Balanced Stratified Cross-Validation in Imbalanced Learning

A gated recurrent unit model based on ultrasound images of dynamic tongue movement for determining the severity of obstructive sleep apnea

Obstructive sleep apnea (OSA) presents as a respiratory disorder characterized by recurrent upper pharyngeal airway collapse during sleep. Dynamic tongue movement (DTM) analysis emerges as a promising avenue for elucidating the pathophysiological underpinnings of OSA, thereby facilitating its diagnosis. Recent endeavors have utilized artificial intelligence techniques to categorize OSA severity leveraging electrocardiography and blood oxygen saturation data. Nonetheless, the integration of ultrasound (US) imaging of the tongue remains largely untapped in the development of machine learning models aimed at determining the severity of OSA. This study endeavors to bridge this gap by capturing US images of DTM dynamics during wakefulness, encompassing transitions from normal breathing (NB) to the performance of the Müller maneuver (MM) in a cohort of 53 patients. Leveraging the modified optical flow method (MOFM), the trajectories of patients' DTM were tracked, facililtating the extraction of 27 parameters vital for model training. These parameters encompassed nine-point lateral movement, nine-point axial movement, and nine-point total displacement of the tongue, resulting in a dataset of 186,030 samples. The gated recurrent unit (GRU) method, renowned for its efficacy in motion tracking, was employed for model development in this study. Validation of the developed model was conducted via stratified k-fold cross-validation (SCV). The systems' overall performance in classifying OSA severity, as quantified by mean accuracy (MA), yielded a value of 43.49%. This pilot investigation marks an exploratory endeavor into the utilization of artificial intelligence for the classification of OSA severity based on US images and dynamic movement patterns. This novel model holds potential to assist clinicians in categorizing OSA severity and guiding the selection of pertinent treatment modalities tailored to the individual needs of patients afflicted with OSA.

Time-Dependent Deep Learning Prediction of Multiple Sclerosis Disability

The majority of deep learning models in medical image analysis concentrate on single snapshot timepoint circumstances, such as the identification of current pathology on a given image or volume. This is often in contrast to the diagnostic methodology in radiology where presumed pathologic findings are correlated to prior studies and subsequent changes over time. For multiple sclerosis (MS), the current body of literature describes various forms of lesion segmentation with few studies analyzing disability progression over time. For the purpose of longitudinal time-dependent analysis, we propose a combinatorial analysis of a video vision transformer (ViViT) benchmarked against traditional recurrent neural network of Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) architectures and a hybrid Vision Transformer-LSTM (ViT-LSTM) to predict long-term disability based upon the Extended Disability Severity Score (EDSS). The patient cohort was procured from a two-site institution with 703 patients' multisequence, contrast-enhanced MRIs of the cervical spine between the years 2002 and 2023. Following a competitive performance analysis, a VGG-16-based CNN-LSTM was compared to ViViT with an ablation analysis to determine time-dependency of the models. The VGG16-LSTM predicted trinary classification of EDSS score in 6 years with 0.74 AUC versus the ViViT with 0.84 AUC (p-value < 0.001 per 5 × 2 cross-validation F-test) on an 80:20 hold-out testing split. However, the VGG16-LSTM outperformed ViViT when patients with only 2 years of MRIs (n = 94) (0.75 AUC versus 0.72 AUC, respectively). Exact EDSS classification was investigated for both models using both classification and regression strategies but showed collectively worse performance. Our experimental results demonstrate the ability of time-dependent deep learning models to predict disability in MS using trinary stratification of disability, mimicking clinical practice. Further work includes external validation and subsequent observational clinical trials.

Feature group partitioning: an approach for depression severity prediction with class balancing using machine learning algorithms

In contemporary society, depression has emerged as a prominent mental disorder that exhibits exponential growth and exerts a substantial influence on premature mortality. Although numerous research applied machine learning methods to forecast signs of depression. Nevertheless, only a limited number of research have taken into account the severity level as a multiclass variable. Besides, maintaining the equality of data distribution among all the classes rarely happens in practical communities. So, the inevitable class imbalance for multiple variables is considered a substantial challenge in this domain. Furthermore, this research emphasizes the significance of addressing class imbalance issues in the context of multiple classes. We introduced a new approach Feature group partitioning (FGP) in the data preprocessing phase which effectively reduces the dimensionality of features to a minimum. This study utilized synthetic oversampling techniques, specifically Synthetic Minority Over-sampling Technique (SMOTE) and Adaptive Synthetic (ADASYN), for class balancing. The dataset used in this research was collected from university students by administering the Burn Depression Checklist (BDC). For methodological modifications, we implemented heterogeneous ensemble learning stacking, homogeneous ensemble bagging, and five distinct supervised machine learning algorithms. The issue of overfitting was mitigated by evaluating the accuracy of the training, validation, and testing datasets. To justify the effectiveness of the prediction models, balanced accuracy, sensitivity, specificity, precision, and f1-score indices are used. Overall, comprehensive analysis demonstrates the discrimination between the Conventional Depression Screening (CDS) and FGP approach. In summary, the results show that the stacking classifier for FGP with SMOTE approach yields the highest balanced accuracy, with a rate of 92.81%. The empirical evidence has demonstrated that the FGP approach, when combined with the SMOTE, able to produce better performance in predicting the severity of depression. Most importantly the optimization of the training time of the FGP approach for all of the classifiers is a significant achievement of this research.

Deep convolutional neural network for weld defect classification in radiographic images

The quality of welds is critical to the safety of structures in construction, so early detection of irregularities is crucial. Advances in machine vision inspection technologies, such as deep learning models, have improved the detection of weld defects. This paper presents a new CNN model based on ResNet50 to classify four types of weld defects in radiographic images: crack, pore, non-penetration, and no defect. Stratified cross-validation, data augmentation, and regularization were used to improve generalization and avoid over-fitting. The model was tested on three datasets, RIAWELC, GDXray, and a private dataset of low image quality, obtaining an accuracy of 98.75 %, 90.255 %, and 75.83 %, respectively. The model proposed in this paper achieves high accuracies on different datasets and constitutes a valuable tool to improve the efficiency and effectiveness of quality control processes in the welding industry. Moreover, experimental tests show that the proposed approach performs well on even low-resolution images.

MGMT ProFWise: Unlocking a New Application for Combined Feature Selection and the Rank-Based Weighting Method to Link MGMT Methylation Status to Serum Protein Expression in Patients with Glioblastoma

Glioblastoma (GBM) is a fatal brain tumor with limited treatment options. O6-methylguanine-DNA-methyltransferase (MGMT) promoter methylation status is the central molecular biomarker linked to both the response to temozolomide, the standard chemotherapy drug employed for GBM, and to patient survival. However, MGMT status is captured on tumor tissue which, given the difficulty in acquisition, limits the use of this molecular feature for treatment monitoring. MGMT protein expression levels may offer additional insights into the mechanistic understanding of MGMT but, currently, they correlate poorly to promoter methylation. The difficulty of acquiring tumor tissue for MGMT testing drives the need for non-invasive methods to predict MGMT status. Feature selection aims to identify the most informative features to build accurate and interpretable prediction models. This study explores the new application of a combined feature selection (i.e., LASSO and mRMR) and the rank-based weighting method (i.e., MGMT ProFWise) to non-invasively link MGMT promoter methylation status and serum protein expression in patients with GBM. Our method provides promising results, reducing dimensionality (by more than 95%) when employed on two large-scale proteomic datasets (7k SomaScan® panel and CPTAC) for all our analyses. The computational results indicate that the proposed approach provides 14 shared serum biomarkers that may be helpful for diagnostic, prognostic, and/or predictive operations for GBM-related processes, given further validation.

Deep Learning Powered Identification of Differentiated Early Mesoderm Cells from Pluripotent Stem Cells

Pluripotent stem cells can be differentiated into all three germ-layers including ecto-, endo-, and mesoderm in vitro. However, the early identification and rapid characterization of each germ-layer in response to chemical and physical induction of differentiation is limited. This is a long-standing issue for rapid and high-throughput screening to determine lineage specification efficiency. Here, we present deep learning (DL) methodologies for predicting and classifying early mesoderm cells differentiated from embryoid bodies (EBs) based on cellular and nuclear morphologies. Using a transgenic murine embryonic stem cell (mESC) line, namely OGTR1, we validated the upregulation of mesodermal genes (Brachyury (T): DsRed) in cells derived from EBs for the deep learning model training. Cells were classified into mesodermal and non-mesodermal (representing endo- and ectoderm) classes using a convolutional neural network (CNN) model called InceptionV3 which achieved a very high classification accuracy of 97% for phase images and 90% for nuclei images. In addition, we also performed image segmentation using an Attention U-Net CNN and obtained a mean intersection over union of 61% and 69% for phase-contrast and nuclear images, respectively. This work highlights the potential of integrating cell culture, imaging technologies, and deep learning methodologies in identifying lineage specification, thus contributing to the advancements in regenerative medicine. Collectively, our trained deep learning models can predict the mesoderm cells with high accuracy based on cellular and nuclear morphologies.

Prediction of Prostate Cancer Risk Stratification Based on A Nonlinear Transformation Stacking Learning Strategy

PURPOSE

Prostate cancer (PCa) is an epithelial malignancy that originates in the prostate gland and is generally categorized into low, intermediate, and high-risk groups. The primary diagnostic indicator for PCa is the measurement of serum prostate-specific antigen (PSA) values. However, reliance on PSA levels can result in false positives, leading to unnecessary biopsies and an increased risk of invasive injuries. Therefore, it is imperative to develop an efficient and accurate method for PCa risk stratification. Many recent studies on PCa risk stratification based on clinical data have employed a binary classification, distinguishing between low to intermediate and high risk. In this paper, we propose a novel machine learning (ML) approach utilizing a stacking learning strategy for predicting the tripartite risk stratification of PCa.

METHODS

Clinical records, featuring attributes selected using the lasso method, were utilized with 5 ML classifiers. The outputs of these classifiers underwent transformation by various nonlinear transformers and were then concatenated with the lasso-selected features, resulting in a set of new features. A stacking learning strategy, integrating different ML classifiers, was developed based on these new features.

RESULTS

Our proposed approach demonstrated superior performance, achieving an accuracy of 0.83 and an area under the receiver operating characteristic curve value of 0.88 in a dataset comprising 197 PCa patients with 42 clinical characteristics.

CONCLUSION

This study aimed to improve clinicians' ability to rapidly assess PCa risk stratification while reducing the burden on patients. This was achieved by using artificial intelligence-related technologies as an auxiliary method for diagnosing PCa.

Time-domain heart rate variability features for automatic congestive heart failure prediction

AIMS

Heart failure is a serious condition that often goes undiagnosed in primary care due to the lack of reliable diagnostic tools and the similarity of its symptoms with other diseases. Non-invasive monitoring of heart rate variability (HRV), which reflects the activity of the autonomic nervous system, could offer a novel and accurate way to detect and manage heart failure patients. This study aimed to assess the feasibility of using machine learning techniques on HRV data as a non-invasive biomarker to classify healthy adults and those with heart failure.

METHODS AND RESULTS

We used digitized electrocardiogram recordings from 54 adults with normal sinus rhythm and 44 adults categorized into New York Heart Association classes 1, 2, and 3, suffering from congestive heart failure. All recordings were sourced from the PhysioNet database. Following data pre-processing, we performed time-domain HRV analysis on all individual recordings, including root mean square of the successive difference in adjacent RR interval (RRi) (RMSSD), the standard deviation of RRi (SDNN, the NN stands for natural or sinus intervals), the standard deviation of the successive differences between successive RRi (SDSD), the number or percentage of RRi longer than 50 ms (NN50 and pNN50), and the average value of RRi [mean RR interval (mRRi)]. In our experimental classification performance evaluation, on the computed HRV parameters, we optimized hyperparameters and performed five-fold cross-validation using four machine learning classification algorithms: support vector machine, k-nearest neighbour (KNN), naïve Bayes, and decision tree (DT). We evaluated the prediction accuracy of these models using performance criteria, namely, precision, recall, specificity, F1 score, and overall accuracy. For added insight, we also presented receiver operating characteristic (ROC) plots and area under the ROC curve (AUC) values. The overall best performance accuracy of 77% was achieved when KNN and DT were trained on computed HRV parameters with a 5 min time window. KNN obtained an AUC of 0.77, while DT attained 0.78. Additionally, in the classification of severe congestive heart failure, KNN and DT had the best accuracy of 91%, with KNN achieving an AUC of 0.88 and DT obtaining 0.92.

CONCLUSIONS

The results show that HRV can accurately predict severe congestive heart failure. The findings of this study could inform the use of machine learning approaches on non-invasive HRV, to screen congestive heart failure individuals in primary care.

Digital image analysis and machine learning-assisted prediction of neoadjuvant chemotherapy response in triple-negative breast cancer

BACKGROUND

Pathological complete response (pCR) is associated with favorable prognosis in patients with triple-negative breast cancer (TNBC). However, only 30-40% of TNBC patients treated with neoadjuvant chemotherapy (NAC) show pCR, while the remaining 60-70% show residual disease (RD). The role of the tumor microenvironment in NAC response in patients with TNBC remains unclear. In this study, we developed a machine learning-based two-step pipeline to distinguish between various histological components in hematoxylin and eosin (H&E)-stained whole slide images (WSIs) of TNBC tissue biopsies and to identify histological features that can predict NAC response.

METHODS

H&E-stained WSIs of treatment-naïve biopsies from 85 patients (51 with pCR and 34 with RD) of the model development cohort and 79 patients (41 with pCR and 38 with RD) of the validation cohort were separated through a stratified eightfold cross-validation strategy for the first step and leave-one-out cross-validation strategy for the second step. A tile-level histology label prediction pipeline and four machine-learning classifiers were used to analyze 468,043 tiles of WSIs. The best-trained classifier used 55 texture features from each tile to produce a probability profile during testing. The predicted histology classes were used to generate a histology classification map of the spatial distributions of different tissue regions. A patient-level NAC response prediction pipeline was trained with features derived from paired histology classification maps. The top graph-based features capturing the relevant spatial information across the different histological classes were provided to the radial basis function kernel support vector machine (rbfSVM) classifier for NAC treatment response prediction.

RESULTS

The tile-level prediction pipeline achieved 86.72% accuracy for histology class classification, while the patient-level pipeline achieved 83.53% NAC response (pCR vs. RD) prediction accuracy of the model development cohort. The model was validated with an independent cohort with tile histology validation accuracy of 83.59% and NAC prediction accuracy of 81.01%. The histological class pairs with the strongest NAC response predictive ability were tumor and tumor tumor-infiltrating lymphocytes for pCR and microvessel density and polyploid giant cancer cells for RD.

CONCLUSION

Our machine learning pipeline can robustly identify clinically relevant histological classes that predict NAC response in TNBC patients and may help guide patient selection for NAC treatment.

A Novel CNN Model for Classification of Chinese Historical Calligraphy Styles in Regular Script Font

Chinese calligraphy, revered globally for its therapeutic and mindfulness benefits, encompasses styles such as regular (Kai Shu), running (Xing Shu), official (Li Shu), and cursive (Cao Shu) scripts. Beginners often start with the regular script, advancing to more intricate styles like cursive. Each style, marked by unique historical calligraphy contributions, requires learners to discern distinct nuances. The integration of AI in calligraphy analysis, collection, recognition, and classification is pivotal. This study introduces an innovative convolutional neural network (CNN) architecture, pioneering the application of CNN in the classification of Chinese calligraphy. Focusing on the four principal calligraphy styles from the Tang dynasty (690-907 A.D.), this research spotlights the era when the traditional regular script font (Kai Shu) was refined. A comprehensive dataset of 8282 samples from these calligraphers, representing the zenith of regular style, was compiled for CNN training and testing. The model distinguishes personal styles for classification, showing superior performance over existing networks. Achieving 89.5-96.2% accuracy in calligraphy classification, our approach underscores the significance of CNN in the categorization of both font and artistic styles. This research paves the way for advanced studies in Chinese calligraphy and its cultural implications.

Multimodality radiomics prediction of radiotherapy-induced the early proctitis and cystitis in rectal cancer patients: a machine learning study

Purpose.This study aims to predict radiotherapy-induced rectal and bladder toxicity using computed tomography (CT) and magnetic resonance imaging (MRI) radiomics features in combination with clinical and dosimetric features in rectal cancer patients.Methods.A total of sixty-three patients with locally advanced rectal cancer who underwent three-dimensional conformal radiation therapy (3D-CRT) were included in this study. Radiomics features were extracted from the rectum and bladder walls in pretreatment CT and MR-T2W-weighted images. Feature selection was performed using various methods, including Least Absolute Shrinkage and Selection Operator (Lasso), Minimum Redundancy Maximum Relevance (MRMR), Chi-square (Chi2), Analysis of Variance (ANOVA), Recursive Feature Elimination (RFE), and SelectPercentile. Predictive modeling was carried out using machine learning algorithms, such as K-nearest neighbor (KNN), Support Vector Machine (SVM), Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Naive Bayes (NB), Gradient Boosting (XGB), and Linear Discriminant Analysis (LDA). The impact of the Laplacian of Gaussian (LoG) filter was investigated with sigma values ranging from 0.5 to 2. Model performance was evaluated in terms of the area under the receiver operating characteristic curve (AUC), accuracy, precision, sensitivity, and specificity.Results.A total of 479 radiomics features were extracted, and 59 features were selected. The pre-MRI T2W model exhibited the highest predictive performance with an AUC: 91.0/96.57%, accuracy: 90.38/96.92%, precision: 90.0/97.14%, sensitivity: 93.33/96.50%, and specificity: 88.09/97.14%. These results were achieved with both original image and LoG filter (sigma = 0.5-1.5) based on LDA/DT-RF classifiers for proctitis and cystitis, respectively. Furthermore, for the CT data, AUC: 90.71/96.0%, accuracy: 90.0/96.92%, precision: 88.14/97.14%, sensitivity: 93.0/96.0%, and specificity: 88.09/97.14% were acquired. The highest values were achieved using XGB/DT-XGB classifiers for proctitis and cystitis with LoG filter (sigma = 2)/LoG filter (sigma = 0.5-2), respectively. MRMR/RFE-Chi2 feature selection methods demonstrated the best performance for proctitis and cystitis in the pre-MRI T2W model. MRMR/MRMR-Lasso yielded the highest model performance for CT.Conclusion.Radiomics features extracted from pretreatment CT and MR images can effectively predict radiation-induced proctitis and cystitis. The study found that LDA, DT, RF, and XGB classifiers, combined with MRMR, RFE, Chi2, and Lasso feature selection algorithms, along with the LoG filter, offer strong predictive performance. With the inclusion of a larger training dataset, these models can be valuable tools for personalized radiotherapy decision-making.

Digital image analysis and machine learning-assisted prediction of neoadjuvant chemotherapy response in triple-negative breast cancer

Background

Pathological complete response (pCR) is associated with favorable prognosis in patients with triple-negative breast cancer (TNBC). However, only 30-40% of TNBC patients treated with neoadjuvant chemotherapy (NAC) show pCR, while the remaining 60-70% show residual disease (RD). The role of the tumor microenvironment (TME) in NAC response in patients with TNBC remains unclear. In this study, we developed a machine learning-based two-step pipeline to distinguish between various histological components in hematoxylin and eosin (H&E)-stained whole slide images (WSIs) of TNBC tissue biopsies and to identify histological features that can predict NAC response.

Methods

H&E-stained WSIs of treatment-naïve biopsies from 85 patients (51 with pCR and 34 with RD) were separated through a stratified 8-fold cross validation strategy for the first step and leave one out cross validation strategy for the second step. A tile-level histology label prediction pipeline and four machine learning classifiers were used to analyze 468,043 tiles of WSIs. The best-trained classifier used 55 texture features from each tile to produce a probability profile during testing. The predicted histology classes were used to generate a histology classification map of the spatial distributions of different tissue regions. A patient-level NAC response prediction pipeline was trained with features derived from paired histology classification maps. The top graph-based features capturing the relevant spatial information across the different histological classes were provided to the radial basis function kernel support vector machine (rbfSVM) classifier for NAC treatment response prediction.

Results

The tile-level prediction pipeline achieved 86.72% accuracy for histology class classification, while the patient-level pipeline achieved 83.53% NAC response (pCR vs. RD) prediction accuracy. The histological class pairs with the strongest NAC response predictive ability were tumor and tumor tumor-infiltrating lymphocytes for pCR and microvessel density and polyploid giant cancer cells for RD.

Conclusion

Our machine learning pipeline can robustly identify clinically relevant histological classes that predict NAC response in TNBC patients and may help guide patient selection for NAC treatment.

Explainable CAD System for Classification of Acute Lymphoblastic Leukemia Based on a Robust White Blood Cell Segmentation

Leukemia is a significant health challenge, with high incidence and mortality rates. Computer-aided diagnosis (CAD) has emerged as a promising approach. However, deep-learning methods suffer from the "black box problem", leading to unreliable diagnoses. This research proposes an Explainable AI (XAI) Leukemia classification method that addresses this issue by incorporating a robust White Blood Cell (WBC) nuclei segmentation as a hard attention mechanism. The segmentation of WBC is achieved by combining image processing and U-Net techniques, resulting in improved overall performance. The segmented images are fed into modified ResNet-50 models, where the MLP classifier, activation functions, and training scheme have been tested for leukemia subtype classification. Additionally, we add visual explainability and feature space analysis techniques to offer an interpretable classification. Our segmentation algorithm achieves an Intersection over Union (IoU) of 0.91, in six databases. Furthermore, the deep-learning classifier achieves an accuracy of 99.9% on testing. The Grad CAM methods and clustering space analysis confirm improved network focus when classifying segmented images compared to non-segmented images. Overall, the proposed visual explainable CAD system has the potential to assist physicians in diagnosing leukemia and improving patient outcomes.