Landslide susceptibility assessment using deep learning considering unbalanced samples distribution

Landslide susceptibility assessment (LSA) is fundamental for managing landslide geological disasters. This study presents a deep learning approach (DNN-MSFM) designed to enhance LSA modeling, particularly addressing limitations caused by the unbalanced distribution of data samples in applied datasets. DNN-MSFM approach combines a deep neural network (DNN) and a mean squared false misclassification loss function (MSFM) to handle unbalanced samples from the algorithmic perspective. The model's performance was evaluated on an unbalanced dataset containing mapping units' records of 293 landslide samples and 653 non-landslide samples from the Baota District, China. Its effectiveness was assessed through statistical metrics and compared against DNN and Support Vector Machine (SVM) basic models. The results demonstrated a significant performance enhancement from the DNN-MSFM (OverallAccuracy = 0.889 and area under the receiver operating characteristic curve (AUC) = 0.84), indicating its effectiveness in learning the underlying landslide susceptibility features and demonstrating its ability to provide improved predictions even in areas with unbalanced landslide samples. Moreover, the study emphasizes the importance of considering balanced loss functions in training DNN under various imbalance degrees and contributes to expanding the applicability of DNN in LSA modeling. Also, this study builds a foundation for further enhancements of deep learning methods for geological disaster assessments.

Stacked deep learning approach for efficient SARS-CoV-2 detection in blood samples

Identifying COVID-19 through blood sample analysis is crucial in managing the disease and improving patient outcomes. Despite its advantages, the current test demands certified laboratories, expensive equipment, trained personnel, and 3-4 h for results, with a notable false-negative rate of 15%-20%. This study proposes a stacked deep-learning approach for detecting COVID-19 in blood samples to distinguish uninfected individuals from those infected with the virus. Three stacked deep learning architectures, namely the StackMean, StackMax, and StackRF algorithms, are introduced to improve the detection quality of single deep learning models. To counter the class imbalance phenomenon in the training data, the Synthetic Minority Oversampling Technique (SMOTE) algorithm is also implemented, resulting in increased specificity and sensitivity. The efficacy of the methods is assessed by utilizing blood samples obtained from hospitals in Brazil and Italy. Results revealed that the StackMax method greatly boosted the deep learning and traditional machine learning methods' capability to distinguish COVID-19-positive cases from normal cases, while SMOTE increased the specificity and sensitivity of the stacked models. Hypothesis testing is performed to determine if there is a significant statistical difference in the performance between the compared detection methods. Additionally, the significance of blood sample features in identifying COVID-19 is analyzed using the XGBoost (eXtreme Gradient Boosting) technique for feature importance identification. Overall, this methodology could potentially enhance the timely and precise identification of COVID-19 in blood samples.

5-Year progression prediction of endplate defects: Utilizing the EDPP-Flow convolutional neural network based on unbalanced data

Background

Lumbar disc degeneration (LDD) is considered as one of the main causes of low back pain. For clinical diagnosis of LDD, magnetic resonance imaging (MRI) is commonly used. Schmorl's node, high intensity zone (HIZ), Modic changes, and other MRI biomarkers of intervertebral disc (IVD) degeneration are also associated with low back pain. However, the progression and natural history of these features are unclear and there is limited predictive capacity with MRI.

Purpose

We aim to establish and validate a deep learning pipeline, EDPP-Flow, for the 5-year progression prediction of Schmorl's node, HIZ, and Modic changes, based on clinical MRIs.

Materials and methods

An MRI dataset developed on 1152 volunteers was used in this study. For each volunteer, two MRI scans, at baseline and 5-year follow-up, were collected and pathology labels were annotated as present or absent (with/without pathology) by two specialists with over 10 years of clinical experience. Our pipeline contained the published MRI-SegFlow and state-of-the-art convolutional neural network for progression prediction of endplate defects. The label distribution of the dataset is unbalanced, where the number of present samples was much smaller than absent samples. The resampling and data augmentation strategies were adopted to increase the number of present samples in the training process and balance the influence of different samples on the model, which can improve the prediction accuracy.

Results

Our pipeline achieved high weighted accuracy, sensitivity, and specificity for progression prediction of Schmorl's node (89.46 ± 3.71%, 89.19 ± 2.70%, 89.72 ± 2.42%), HIZ (91.75 ± 2.48%, 93.07 ± 3.96%, 90.43 ± 2.51%), and Modic changes (87.51 ± 2.23%, 87.93 ± 1.72%, 87.10 ± 1.99%), on the unbalanced dataset (present sample's percentages of the 3 pathologies above were 4.3%, 11.7%, and 6.7%).

Conclusion

We developed and validated a deep learning pipeline, for the progression prediction of endplate defects, which showed high prediction accuracy on unbalanced data. The method has significant potential for clinical implementation.

Process monitoring for quality - A multiple classifier system for highly unbalanced data

In big data-based analyses, because of hyper-dimensional feature spaces, there has been no previous distinction between machine learning algorithms (MLAs). Therefore, multiple diverse algorithms should be included in the analysis to develop an adequate model for detecting/recognizing patterns exhibited by classes. If multiple classifiers are developed, the next natural step is to determine whether the prediction benchmark set by the top performer can be improved by combining them. In this context, multiple classifier systems (MCSs) are powerful solutions for difficult pattern recognition problems because they usually outperform the best individual classifier, and their diversity tends to improve resilience and robustness to high-dimensional and noisy data. To design an MCS, an appropriate fusion method is required to optimally combine the individual classifiers and determine the final decision. Process monitoring for quality is a Quality 4.0 initiative aimed at defect detection via binary classification. Because most mature organizations have merged traditional quality philosophies, their processes generate only a few defects per million of opportunities. Therefore, manufacturing data sets for binary classification of quality tends to be highly/ultra-unbalanced. Detecting these rare quality events is one of the most relevant intellectual challenges posed to the fourth industrial revolution, Industry 4.0 (I 4.0). A new MCS aimed at analyzing these data structures is presented. It is based on eight well-known MLAs, an ad hoc fitness function, and a novel meta-learning algorithm. For predicting the final quality class, this algorithm considers the prediction from a set of classifiers as input and determines which classifiers are reliable and which are not. Finally, to demonstrate the superiority of the MLAs over extensively used fusion rules, multiple publicly available data sets are analyzed.

CWLy-RF: A novel approach for identifying cell wall lyases based on random forest classifier

Drug resistance of pathogenic bacteria has become increasingly serious due to the abuse of antibiotics in recent years. Researchers have found that cell wall lyases are effective antibacterial agents that can specifically recognize target bacteria and degrade bacterial peptidoglycan. Traditional wet experiments are usually expensive, time-consuming and laborious for the identification of lyases. Therefore, there is an urgent need to develop prediction tools based on computer methods to identify lyases quickly and accurately. In this paper, a new predictor, CWLy-RF, is proposed based on the random forest (RF) algorithm to identify cell wall lyases. In this method, we combined three features, namely, 400D, 188D and the composition of k-spaced amino acid group pairs, using mixed-feature representation methods. Afterward, we improved the feature representation ability with the selected top 100 features by using the information gain method and trained a predictive model using RF. The constructed prediction model is evaluated by using 10-fold cross-validation. The accuracy obtained was 96.09%, the AUC was 0.993, the MCC was 0.922, the sensitivity was 94.92%, and the specificity was 97.32%. We have proved that the proposed predictor CWLy-RF is superior to other latest models, and it will hopefully become an effective and useful tool for identifying lyases.

Constraining PERMANOVA and LDM to within-set comparisons by projection improves the efficiency of analyses of matched sets of microbiome data

BACKGROUND

Matched-set data arise frequently in microbiome studies. For example, we may collect pre- and post-treatment samples from a set of individuals, or use important confounding variables to match data from case participants to one or more control participants. Thus, there is a need for statistical methods for data comprised of matched sets, to test hypotheses against traits of interest (e.g., clinical outcomes or environmental factors) at the community level and/or the operational taxonomic unit (OTU) level. Optimally, these methods should accommodate complex data such as those with unequal sample sizes across sets, confounders varying within sets, and continuous traits of interest.

METHODS

PERMANOVA is a commonly used distance-based method for testing hypotheses at the community level. We have also developed the linear decomposition model (LDM) that unifies the community-level and OTU-level tests into one framework. Here we present a new strategy that can be used with both PERMANOVA and the LDM for analyzing matched-set data. We propose to include an indicator variable for each set as covariates, so as to constrain comparisons between samples within a set, and also permute traits within each set, which can account for exchangeable sample correlations. The flexible nature of PERMANOVA and the LDM allows discrete or continuous traits or interactions to be tested, within-set confounders to be adjusted, and unbalanced data to be fully exploited.

RESULTS

Our simulations indicate that our proposed strategy outperformed alternative strategies, including the commonly used one that utilizes restricted permutation only, in a wide range of scenarios. Using simulation, we also explored optimal designs for matched-set studies. The flexibility of PERMANOVA and the LDM for a variety of matched-set microbiome data is illustrated by the analysis of data from two real studies.

CONCLUSIONS

Including set indicator variables and permuting within sets when analyzing matched-set data with PERMANOVA or the LDM is a strategy that performs well and is capable of handling the complex data structures that frequently occur in microbiome studies. Video Abstract.

A two-stage modeling approach for breast cancer survivability prediction

BACKGROUND

Despite the increasing number of studies in breast cancer survival prediction, there is little attention put toward deceased patients and their survival lengths. Moreover, developing a model that is both accurate and interpretable remains a challenge.

OBJECTIVE

This paper proposes a two-stage data analytic framework, where Stage I classifies the survival and deceased statuses and Stage II predicts the number of survival months for deceased females with cancer. Since medical data are not entirely clean nor prepared for model development, we aim to show that data preparation can strengthen a simple Generalized Linear Model (GLM)¹ to predict as accurate as the complex models like Extreme Gradient Boosting (XGB)² and Multilayer Perceptron based on Artificial Neural Networks (MLP-ANNs)³ in both stages.

METHODS

In Stage I, we use recent Surveillance, Epidemiology, and End Results (SEER)⁴ data from 2004 to 2016 to predict short term survival statuses from 6-months to 3-years with 6 month increments. Synthetic Minority Over-sampling Technique (SMOTE),⁵ Relocating Safe-Level SMOTE (RSLS)⁶, Adaptive Synthetic (ADASYN)⁷ re-sampling techniques, Least Absolute Shrinkage and Selection Operator (LASSO)⁸ and Random Forest (RF)⁹ feature selection methods along with integer and one-hot encoding are combined with the three popular data mining methods: GLM, XGB, and MLP. In Stage II, we predict the number of survival months for patients who are correctly predicted as deceased within 3-years. Again, we employ GLM, XGB, and MLP for regression along with LASSO and RF for feature selection and one-hot encoding to encode the categorical features.

RESULTS

We obtain Area Under the Receiver Operating Characteristic Curve (AUC)¹⁰ values of 0.900, 0.898, 0.877, 0.852, 0.852, and 0.858 for 6-month, 1-, 1.5-, 2-, 2.5, and 3-year survival time-points, respectively, using OneHotEncoding-GLM-LASSO-ADASYN. We use the change in the Odds Ratio values in GLM to manifest the impact of individual categorical levels and numerical features on the odds of death. In Stage II, we obtain Mean Absolute Error (MAE)¹¹ of 7.960 months using OneHotEncoding-GLM-LASSO when predicting the number of survival months for deceased patients. We present the top contributing features and their coefficient values to illustrate how the presence of each feature alters the predicted number of survival months.

CONCLUSION

To the best of our knowledge, this is the first study that implements both breast cancer survival classification and regression in a two-stage approach. All data-driven findings are presented in order to assist clinicians make better care decisions using GLM, an interpretable and computationally efficient method that predicts survival status and survival lengths for deceased patients, to help foster human and machine interactions.

Boundary loss for highly unbalanced segmentation

Widely used loss functions for CNN segmentation, e.g., Dice or cross-entropy, are based on integrals over the segmentation regions. Unfortunately, for highly unbalanced segmentations, such regional summations have values that differ by several orders of magnitude across classes, which affects training performance and stability. We propose a boundary loss, which takes the form of a distance metric on the space of contours, not regions. This can mitigate the difficulties of highly unbalanced problems because it uses integrals over the interface between regions instead of unbalanced integrals over the regions. Furthermore, a boundary loss complements regional information. Inspired by graph-based optimization techniques for computing active-contour flows, we express a non-symmetric L₂ distance on the space of contours as a regional integral, which avoids completely local differential computations involving contour points. This yields a boundary loss expressed with the regional softmax probability outputs of the network, which can be easily combined with standard regional losses and implemented with any existing deep network architecture for N-D segmentation. We report comprehensive evaluations and comparisons on different unbalanced problems, showing that our boundary loss can yield significant increases in performances while improving training stability. Our code is publicly available¹.

A Review of Microarray Datasets: Where to Find Them and Specific Characteristics

The advent of DNA microarray datasets has stimulated a new line of research both in bioinformatics and in machine learning. This type of data is used to collect information from tissue and cell samples regarding gene expression differences that could be useful for disease diagnosis or for distinguishing specific types of tumor. Microarray data classification is a difficult challenge for machine learning researchers due to its high number of features and the small sample sizes. This chapter is devoted to reviewing the microarray databases most frequently used in the literature. We also make the interested reader aware of the problematic of data characteristics in this domain, such as the imbalance of the data, their complexity, and the so-called dataset shift.

Deep Convolutional Neural Networks for the Prediction of Molecular Properties: Challenges and Opportunities Connected to the Data

We present a flexible deep convolutional neural network method for the analysis of arbitrary sized graph structures representing molecules. This method, which makes use of the Lipinski RDKit module, an open-source cheminformatics software, enables the incorporation of any global molecular (such as molecular charge and molecular weight) and local (such as atom hybridization and bond orders) information. In this paper, we show that this method significantly outperforms another recently proposed method based on deep convolutional neural networks on several datasets that are studied. Several best practices for training deep convolutional neural networks on chemical datasets are also highlighted within the article, such as how to select the information to be included in the model, how to prevent overfitting and how unbalanced classes in the data can be handled.

Two-way model with random cell sizes

We consider inference for row effects in the presence of possible interactions in a two-way fixed effects model when the numbers of observations are themselves random variables. Let N_ij be the number of observations in the (i, j) cell, π_ij be the probability that a particular observation is in that cell and μ_ij be the expected value of an observation in that cell. We assume that the {N_ij } have a joint multinomial distribution with parameters n and {π_ij }. Then μ̄_i . = Σ _jπ_ijμ_ij /Σ _jπ_ij is the expected value of a randomly chosen observation in the ith row. Hence, we consider testing that the μ̄_i . are equal. With the {π_ij } unknown, there is no obvious sum of squares and F-ratio computed by the widely available statistical packages for testing this hypothesis. Let Ȳ_i .. be the sample mean of the observations in the ith row. We show that Ȳ_i .. is an MLE of μ̄_i ., is consistent and is conditionally unbiased. We then find the asymptotic joint distribution of the Ȳ_i .. and use it to construct a sensible asymptotic size α test of the equality of the μ̄_i . and asymptotic simultaneous (1 - α) confidence intervals for contrasts in the μ̄_i ..