Class Imbalance Problem Research Articles

HRCUNet: Hierarchical Region Contrastive Learning for Segmentation of Breast Tumors in DCE‐MRI

ABSTRACTSegmenting breast tumors from dynamic contrast‐enhanced magnetic resonance images is a critical step in the early detection and diagnosis of breast cancer. However, this task becomes significantly more challenging due to the diverse shapes and sizes of tumors, which make it difficult to establish a unified perception field for modeling them. Moreover, tumor regions are often subtle or imperceptible during early detection, exacerbating the issue of extreme class imbalance. This imbalance can lead to biased training and challenge accurately segmenting tumor regions from the predominant normal tissues. To address these issues, we propose a hierarchical region contrastive learning approach for breast tumor segmentation. Our approach introduces a novel hierarchical region contrastive learning loss function that addresses the class imbalance problem. This loss function encourages the model to create a clear separation between feature embeddings by maximizing the inter‐class margin and minimizing the intra‐class distance across different levels of the feature space. In addition, we design a novel Attention‐based 3D Multi‐scale Feature Fusion Residual Module to explore more granular multi‐scale representations to improve the feature learning ability of tumors. Extensive experiments on two breast DCE‐MRI datasets demonstrate that the proposed algorithm is more competitive against several state‐of‐the‐art approaches under different segmentation metrics.

Ensemble feature selection and tabular data augmentation with generative adversarial networks to enhance cutaneous melanoma identification and interpretability

BackgroundCutaneous melanoma is the most aggressive form of skin cancer, responsible for most skin cancer-related deaths. Recent advances in artificial intelligence, jointly with the availability of public dermoscopy image datasets, have allowed to assist dermatologists in melanoma identification. While image feature extraction holds potential for melanoma detection, it often leads to high-dimensional data. Furthermore, most image datasets present the class imbalance problem, where a few classes have numerous samples, whereas others are under-represented.MethodsIn this paper, we propose to combine ensemble feature selection (FS) methods and data augmentation with the conditional tabular generative adversarial networks (CTGAN) to enhance melanoma identification in imbalanced datasets. We employed dermoscopy images from two public datasets, PH2 and Derm7pt, which contain melanoma and not-melanoma lesions. To capture intrinsic information from skin lesions, we conduct two feature extraction (FE) approaches, including handcrafted and embedding features. For the former, color, geometric and first-, second-, and higher-order texture features were extracted, whereas for the latter, embeddings were obtained using ResNet-based models. To alleviate the high-dimensionality in the FE, ensemble FS with filter methods were used and evaluated. For data augmentation, we conducted a progressive analysis of the imbalance ratio (IR), related to the amount of synthetic samples created, and evaluated the impact on the predictive results. To gain interpretability on predictive models, we used SHAP, bootstrap resampling statistical tests and UMAP visualizations.ResultsThe combination of ensemble FS, CTGAN, and linear models achieved the best predictive results, achieving AUCROC values of 87% (with support vector machine and IR=0.9) and 76% (with LASSO and IR=1.0) for the PH2 and Derm7pt, respectively. We also identified that melanoma lesions were mainly characterized by features related to color, while not-melanoma lesions were characterized by texture features.ConclusionsOur results demonstrate the effectiveness of ensemble FS and synthetic data in the development of models that accurately identify melanoma. This research advances skin lesion analysis, contributing to both melanoma detection and the interpretation of main features for its identification.

A domain generalization network for imbalanced machinery fault diagnosis

Traditional models for Imbalanced Fault Diagnosis (IFD) face challenges in practical applications due to domain shifts caused by varying working conditions and machinery. Domain Generalization (DG) models provide an advantage over traditional approaches by learning class-discriminative and domain-invariant feature representations, allowing them to generalize to unseen target data. However, the scarcity of fault samples relative to healthy ones limits their application in real-world industrial scenarios. In this paper, we propose a Domain Mixed-Enhanced Domain Generalization Network (DEMDGN) that enhances IFD performance by utilizing mixup-based data augmentation and domain-based discrepancy metrics to align feature distributions across multiple heterogeneous source domains. By creating domain-invariant features, DEMDGN allows robust fault diagnosis under varying conditions. Extensive experiments on one marine machinery dataset and two bearing datasets demonstrate that the proposed method effectively addresses class imbalance and domain shift problems, achieving superior diagnostic performance.

LES-YOLO: efficient object detection algorithm used on UAV for traffic monitoring

Abstract The use of UAVs for traffic monitoring greatly facilitates people’s lives. Classical object detection algorithms struggle to balance high speed and accuracy when processing UAV images on edge devices. To solve the problem, the paper introduces an efficient and slim YOLO with low computational overhead, named LES-YOLO. In order to enrich the feature representation of small and medium objects in UAV images, a redesigned backbone is introduced. And C2f combined with Coordinate Attention is used to focus on key features. In order to enrich cross-scale information and reduce feature loss during network transmission, a novel structure called EMS-PAN (Enhanced Multi-Scale PAN) is designed. At the same time, to alleviate the problem of class imbalance, Focal EIoU is used to optimize network loss calculation instead of CIoU. To minimize redundancy and ensure a slim architecture, the P5 layer has been eliminated from the model. And verification experiments show that LES-YOLO without P5 is more efficient and slimmer. LES-YOLO is trained and tested on the VisDrone2019 dataset. Compared with YOLOv8n-p2, mAP@0.5 and Recall has increased by 7.4% and 7%. The number of parameters is reduced by over 50%, from 2.9 M to 1.4 M, but there is a certain degree of increase in FLOPS, reaching 18.8 GFLOPS. However, the overall computational overhead is still small enough. Moreover, compared with YOLOv8s-p2, both the number of parameters and FLOPS are significantly reduced , while the performance is similar . As for real-time, LES-YOLO reaches 138 fps on GPU and a maximum of 78 fps on edge devices of UAV.

A novel generalised extreme value gradient boosting decision tree for the class imbalanced problem in credit scoring

The performance of credit scoring models can be compromised when dealing with imbalanced datasets, where the number of defaulted borrowers is significantly lower than that of non-defaulters. To address this challenge, we propose a gradient boosting decision tree with the generalised extreme value distribution model (GEV-GBDT). Our approach replaces the conventional symmetric logistic sigmoid function with the asymmetric cumulative distribution function of the GEV distribution as the activation function. We derive a novel loss function based on the maximum likelihood estimation of the GEV distribution within the boosting framework. This modification allows the model to focus more on the minority class by emphasising the tail of the response curve, and the shape parameter of the GEV distribution offers flexibility in controlling the model’s emphasis on minority samples. We examine the performance of this approach using four real-life loan datasets. The empirical results show that the GEV-GBDT model achieves superior classification performance compared to other commonly used imbalanced learning methods, including the synthetic minority oversampling technique and the cost-sensitive framework. Furthermore, we conduct performance tests on several datasets with varying imbalance ratios and find that GEV-GBDT performs better on extremely imbalanced datasets.

A hybrid approach of vision transformers and CNNs for detection of ulcerative colitis

Ulcerative Colitis is an Inflammatory Bowel disease caused by a variety of factors that lead to a serious impact on the quality of life of the patients if left untreated. Due to complexities in the identification procedures of this disease, the treatment timeline and quality can be severely affected, leading to further consequences for the sufferer. The difficulties in identification are due to high patients to healthcare professionals ratio. Researchers have proposed variety of machine/deep learning methods for automated detection of ulcerative colitis, however, several challenges exists including class imbalance problem, comprehensive feature extraction and accurate classification. We propose a novel method for accurate detection of ulcerative colitis with augmentation techniques to overcome class imbalance issue, a comprehensive feature vector extraction using custom architecture of Vision Transformer (ViT) and accurate classification using customized Convolutional Neural Network (CNN). We used the TMC-UCM and LIMUC datasets in this research for training and testing of proposed method and achieved accuracy of 90% with AUC-ROC scores of 0.91, 0.81, 0.94, and 0.94 for the endoscopic classes of Mayo 0, Mayo 1, Mayo 2, and Mayo 3 respectively. We have compared the proposed method with existing state of the art methods and conclude that the proposed method outperforms the existing methods.

Automated lesion detection in cotton leaf visuals using deep learning

Cotton is one of the major cash crop in the agriculture led economies across the world. Cotton leaf diseases affects its yield globally. Determining cotton lesions on leaves is difficult when the area is big and the size of lesions is varied. Automated cotton lesion detection is quite useful; however, it is challenging due to fewer disease class, limited size datasets, class imbalance problems, and need of comprehensive evaluation metrics. We propose a novel deep learning based method that augments the data using generative adversarial networks (GANs) to reduce the class imbalance issue and an ensemble-based method that combines the feature vector obtained from the three deep learning architectures including VGG16, Inception V3, and ResNet50. The proposed method offers a more precise, efficient and scalable method for automated detection of diseases of cotton crops. We have implemented the proposed method on publicly available dataset with seven disease and one health classes and have achieved highest accuracy of 95% and F-1 score of 98%. The proposed method performs better than existing state of the art methods.

Aero-engine defect detection by integrating attention and multi-scale features

The structural integrity of aircraft engines is related to flight safety. At present, aircraft engine defect detection based on borescope technology is mainly manual operation. In order to improve the detection accuracy and efficiency, an intelligent aircraft engine defect detection algorithm that integrates attention and multi-scale features is proposed to assist borescope work. First, to address the class imbalance problem of defect samples in the original borescope image, a multi- sample fusion data enhancement method based on geometric transformation and Poisson image editing is used to enrich small sample images and construct a defect dataset. Then, the coordinated attention module (CA) is integrated into the baseline network YOLOv5 to emphasize the extraction of defect features and enhance the network's distinction between defect targets and complex backgrounds. A weighted bidirectional feature pyramid structure (BiFPN) is constructed in the neck network to complete higher-level feature fusion and improve the expression ability of multi-scale targets. Finally, the bounding box regression loss function is defined as the EIOU loss to achieve fast and accurate positioning and identification of defect targets. Experimental results show that the average accuracy of defect detection of the proposed algorithm reaches 89.7%, which is improved compared with the baseline network 6.3%, and the size of the trained model is only 14.0 M. Therefore, the proposed method can effectively detect the main defects of aircraft engines.

Respiratory disease detection in lung auscultation with convolutional neural networks and CVAE augmentation

The main purpose of this work was to investigate the possibility of detecting respiratory diseases in audio recordings of lung auscultation using modern deep learning tools, as well as to explore the possibility of using data augmentation by generating synthetic spectral representations of audio samples. The ICBHI (International Conference on Biomedical and Health Informatics) dataset was used for training, validation and augmentation. The dataset includes lung auscultations of 126 different subjects, there are a total of 920 sounds, of which 810 have signs of chronic diseases, 75 of non-chronic diseases and 35 with no pathology. The stage of data preprocessing includes discretization to 4kHz frequency, as well as filtering of frequency bands that do not carry information value for the task. In the next step, each sample was transformed into a frequency spectrum and Melspectrograms were generated. To solve the problem of class imbalance, the required number of synthetic spectrograms generated by convolutional variation autoencoders was added. At the stage of building the model, the methods of classical convolutional neural networks were used. The quality of the obtained algorithm was evaluated using a 10-fold cross-validation. Also, to assess the generalization of the proposed method, experiments were performed with the split of audio recordings into training and test sets using patient grouping. Qualitative evaluation of the model was performed using sensitivity, specificity, F1-score and Cohen’s kappa. A score of 98.45% F1-score was achieved for the 5-class classification problem which can contribute to the development of ways to synthesize and augment sensitive medical data. In addition, a cons of existing methods in the generalization of the obtained predictions were revealed, which opens the way for further research in the direction of clinical respiratory diseases detection.

A multi-class fundus disease classification system based on an adaptive scale discriminator and hybrid loss

Fundus images are crucial in the observation and detection of ophthalmic diseases. However, detecting multiple ophthalmic diseases from fundus images using deep learning techniques is a complex and challenging task One challenge is the complexity of fundus disease structures, which leads to low detection accuracy. Another challenge is the class imbalance problem common in multi-label image classification, which increases the difficulty of algorithm training and evaluation. To address these issues, this study leverages deep learning to propose an ophthalmic disease classification system. We first employ ResNet50 as the backbone network to extract image features, and then use our designed multi-dimensional attention module and adaptive scale discriminator to enhance the network's ability to detect disease features. During training, we innovatively propose a hybrid loss function method to improve the detection capability on imbalanced data. Finally, we conducted experiments on the ODRI-5K dataset with the proposed classification system. In the test set, our method achieved an AUC of 98.53 and an F1-score of 89.73. This result fully demonstrates the excellent disease classification capability of our method. In summary, the multi-label fundus image disease classification system we proposed exhibits outstanding recognition capability, providing an effective solution for the diagnosis of multi-label fundus image diseases.

From shallows to depths: unveiling hybrid synthetic data modeling for enhanced learning with privacy considerations in naturally imbalanced datasets

Data serves as the foundational element that drives model development and performance in machine learning and deep learning. The learning algorithms are as good as data on which they are trained on. Data constrained environments pose serious threat to the effectiveness of learning algorithms. Data limitations stem from a range of factors, including regulatory restrictions, privacy concerns, and the inherent scarcity of relevant data. This constrained availability of data often leads to class imbalance problem in the context of classification tasks. To address the challenges of limited data, the algorithmic generated data, also known as synthetic data, is gaining significant traction as a cost-effective, readily available, and secure alternative. Synthetic data generation techniques can be employed to enhance dataset size by augmenting data samples and to address class imbalance by increasing the number of minority class instances. These techniques generally fall into two main categories: distance-based shallow models and probability estimation-based deep generative models. Shallow interpolation-based models generate new data points within the local space between existing data points, while deep density estimation based models generate new data by learning the whole distribution of data. . In the context of smaller datasets, deep generative models often struggle to accurately estimate the probability distribution of whole data. To effectively represent the global data distribution, these deep models require initial starting data samples to guide their approximation. This paper examines the potential of integration of shallow and deep generative models in the data generation pipeline for effective synthetic data augmentation which furthers enhanced learning and generalization of downstream tasks. In this work, we present a hybrid approach of tabular data generation involving mixed type data attributes (continuous, discrete) and pay special attention to data imbalance and insufficient data problems. We introduce the Hybrid Data Balancing and Augmentation Approach for Mixed Tabular Data (HDBA-MTD), specifically designed to synthesize samples for underrepresented labels of output class and address issues of insufficient data instances. This approach enhances training data diversity, thereby paying special attention to the downstream classification and generalization performance. Experiments are carried out using benchmark datasets to assess the practicality of the presented hybrid model in real-world scenarios. This work has also attempted to quantify the privacy preservability for real data concerning ethical considerations and data security circumstances. The evaluation and analysis of these experiments show that the present hybrid model performs favorably compared to other current hybrid synthetic data generation methods.

Credit risk assessment method driven by asymmetric loss function

Credit risk assessment is significantly hindered by the problem of class imbalance, and cost-sensitive methods represent an effective strategy to address this issue. However, most algorithms tend to approach the imbalance from a class perspective, overlooking the finer details at the sample level. Moreover, such methods are susceptible to interference from noise and outliers. In response to these challenges, this paper proposes an asymmetric cost-sensitive support vector machine (QTSVM) that combines the quadratic type squared error loss function (QTSELF) with support vector machine (SVM). It not only leverages the asymmetry of the loss function by applying varying penalties based on classification errors but also employs different processing measures for samples from different classes. Additionally, it enhances model robustness by imposing a tiny penalty on noise or outliers. The adaptive moment estimation (Adam) algorithm is employed to optimize QTSVM. Extensive experiments and statistical tests profoundly demonstrate the superior performance of QTSVM.

A multiscale approach for network intrusion detection based on variance–covariance subspace distance and EQL v2

As an important network defense approach, network intrusion detection is mainly used to identify anomaly traffic behavior. However, dominant network intrusion detection approaches are now struggling to identify the complex and variable means of attack, leading to high false alarm rate. Additionally, the feature redundancy and class imbalance problem in the intrusion detection dataset also constrain the performance of detection methods. This paper proposes a multiscale intrusion detection approach based on variance–covariance subspace distance and Equalization Loss v2 (EQL v2). Firstly, the variance–covariance subspace distance is used for feature selection on the preprocessed dataset to determine a set of representative feature subsets that can effectively approximate the original feature space. Secondly, the loss function, EQL v2, is adopted to balance the positive and negative gradients, addressing the class imbalance problem. Finally, a pyramid depthwise separable convolution model is proposed to capture the multiscale information of the traffic, and the convolutional layer in the depthwise convolution is replaced with self-supervised predictive convolutional attention block to compensate for the performance loss caused by the parameter reduction. Extensive experiments demonstrated that the proposed approach exhibits better performance on the three datasets of NSL-KDD, UNSW_NB15, and CIC-IDS-2017, with accuracy rates of 99.19%, 97.81%, and 99.83%, respectively, effectively improve the intrusion detection performance.

Analyzing Resampling Techniques for Addressing the Class Imbalance in NIDS using SVM with Random Forest Feature Selection

The purpose of Network Intrusion Detection Systems (NIDS) is to ensure and protect computer networks from harmful actions. A major concern in NIDS development is the class imbalance problem, i.e., normal traffic dominates the communication data plane more than intrusion attempts. Such a state of affairs can pose certain hazards to the effectiveness of detection algorithms, including those useful for detecting less frequent but still highly dangerous intrusions. This paper aims to utilize resampling techniques to tackle this problem of class imbalance in NIDS using a Support Vector Machine (SVM) classifier alongside utilizing features selected by Random Forest to improve the feature subset selection process. The analysis highlights the combativeness of each sampling method, offering insights into their efficiency and practicality for real-world applications. Four resampling techniques are analyzed. Such techniques include Synthetic Minority Over-sampling Technique (SMOTE), Random Under-sampling (RUS), Random Over-sampling (ROS) and SMOTE with two different combinations i.e., RUS SMOTE and RUS ROS. Feature selection was done using Random Forest, which was improved by Bayesian methods to create subsets of features with feature rankings determined by Cumulative Feature Importance Score (CFIS). The CIDDS-2017 dataset is used for the performance evaluation, and the metrics used include accuracy, precision, recall, F-measure and CPU time. The algorithm that performs best overall in the CFIS feature subsets is SMOTE, and the features that give the best result are selected at the 90% level with 25 features. This subset accomplishes a relative accuracy enhancement of 0.08% than the other approaches. The RUS+ROS technique is also fine but somehow slower than SMOTE. On the other hand, RUS+SMOTE shows relatively poor results although it consumes less time in terms of computational time compared to other methods, giving about 50% of the performance shown by the other methods. This paper's novelty is adapting the RUS method as a standalone test for screening new and potentially contaminated datasets. The standalone RUS method is more efficient in terms of computations; the algorithm returned the best result of 98.13% accuracy at 85% at the CFIS level of 34 features with a computation time of 137.812 s. It is also noted that SMOTE is considered to be proficient among all resampling techniques used for handling the problem of class imbalance in NIDS, vice 90% CFIS feature subset. Future research directions could include using these techniques in different data sets and other machine learning and deep learning methods together with ROC curve analysis to provide useful pointers to NIDS designers on how to select the right data mining tools and strategies for their projects.

Classification of Air Pollutant Index on Data with Outliers and Imbalance Class Problem

The problem of air pollution has become a global issue that has received attention from various countries. Jakarta, Indonesia's capital city, is unavoidable from the same problem. This study will use four parameters of substances PM10, SO2, CO, O3, and nitrogen dioxide to categorize Jakarta's air quality (NO2). The data used is daily data taken from the Air Quality Monitoring Station in Jakarta throughout 2020. The methods used include SVM, Random Forest, Logistic Regression, KNN, CART, and Stacking Algorithm. At the data preparation stage, we found missing values, outliers, and class imbalance problems. Before applying machine learning methods and evaluating accuracy, we used data pre-processing techniques such as the MissForest method, median substitution, and ADASYN. The results prove that the original dataset has a higher accuracy score (0.882 – 0.977) than the balanced dataset (0.829 – 0.976). According to the evaluation results, the Random Forest method has the highest accuracy score for original and balanced datasets. The overall result is better than the identical research, which produces 96.61% accuracy using a neural network. It shows that preprocessing steps such as missing values handling an imbalanced class handling is essential in classification performance.

Feature selection via robust weighted score for high dimensional binary class-imbalanced gene expression data

In this paper, a robust weighted score for unbalanced data (ROWSU) is proposed for selecting the most discriminative features for high dimensional gene expression binary classification with class-imbalance problem. The method addresses one of the most challenging problems of highly skewed class distributions in gene expression datasets that adversely affect the performance of classification algorithms. First, the training dataset is balanced by synthetically generating data points from minority class observations. Second, a minimum subset of genes is selected using a greedy search approach. Third, a novel weighted robust score, where the weights are computed by support vectors, is introduced to obtain a refined set of genes. The highest scoring genes based on this approach are combined with the minimum subset of genes selected by the greedy search approach to form the final set of genes. The novel method ensures the selection of the most discriminative genes, even in the presence of skewed class distribution, thereby improving the performance of the classifiers. The performance of the proposed ROWSU method is evaluated on 7 gene expression datasets. Classification accuracy, sensitivity and F1-score are used as performance metrics to compare the proposed ROWSU algorithm with several other state-of-the-art methods. Boxplots and stability plots are also constructed for a better understanding of the results. The results show that the proposed method outperforms the existing feature selection procedures based on classification performance from k nearest neighbors (kNN) and random forest (RF) classifiers.

A multi-strategy ontology mapping method based on cost-sensitive SVM

As the core of ontology integration, the task of ontology mapping is to find the semantic relationship between ontologies. Nevertheless, most existing ontology mapping methods only rely ontext information to calculate entity similarity, thereby disregarding semantic nuances and necessitating substantial manual intervention. Therefore, this paper introduces an ontology mapping method. Based on the traditional ontology mapping method, the process employs a deep learning model to mine the semantic information of entity concepts, entity properties and ontology structure to obtain the embedding vector. We use the similarity mechanism to calculate the similarity between different embedding vectors, and combine the similarity values obtained from multiple strategy entities into a similarity matrix. The similarity matrix serves as input to the support vector machine (SVM), and the ontology mapping problem is finally transformed into a binary classification problem. However, since the number of mapped pairs is much larger than the number of non-mapped pairs, the number of positive samples in the data set is much smaller than the number of negative samples. Therefore, based on the traditional SVM, the paper adopts cost-sensitive strategy to deal with the class imbalance problem. In comparative evaluations against contemporary ontology mapping techniques, our method demonstrates a noteworthy 5.0% enhancement in recall and a 3.0% improvement in F1 score when tested on both public benchmark datasets and domain-specific datasets.

Metaheuristic-driven space partitioning and ensemble learning for imbalanced classification

Imbalanced classification is a common issue in Machine Learning, particularly when misclassifying minor instances leads to significant costs. In literature, various strategies have been employed to address this problem. These include data-level, algorithm-level, cost-sensitive, and hybrid-level algorithms designed to tackle imbalanced problems. This paper aims to introduce a novel method that simultaneously enhances the ability of classification models to identify patterns more effectively and addresses imbalanced problems while minimizing alterations to the original data distribution. Our proposed framework combines ensemble learning, space partitioning, and the Synthetic Minority Oversampling Technique (SMOTE). This method decomposes the space into balanced sub-spaces and then trains an ensemble classifier based on these sub-spaces using a bagging approach. In the initial step, we develop a Space Partitioning by Metaheuristic algorithm (SPMH) to divide the space into multiple balanced subspaces. In the subsequent step, we present Imbalanced Classification by SPMH (ICSPMH) as a solution to imbalanced class problems. ICSPMH uses SPMH multiple times to divide the space into different sub-spaces, creating various sub-spaces each time. It then trains different classifiers for each portion of the space, creating an ensemble classifier through a bagging technique. To assess the performance of our proposed framework, we selected 44 well-known datasets for comparison with some state-of-the-art approaches. The results demonstrate that ICSPMH outperforms other competent methods and can potentially reduce the oversampling rate to zero. Additionally, an experiment indicated that the choice of metaheuristic algorithm in SPMH does not significantly impact the final performance. The paper also includes a correlation analysis between oversampling rate and final performance, revealing that the framework effectively eliminates imbalanced data problems with minimal changes to the original dataset. In summary, because ICSPMH applies fewer changes in data distribution and sets up local classifiers that improve classification performance, it looks like a promising method for classifying imbalanced datasets.

An Artificial Intelligence-Based Tool for Enhancing Pectoral Muscle Segmentation in Mammograms: Addressing Class Imbalance and Validation Challenges in Automated Breast Cancer Diagnosis.

Breast cancer remains a major health concern worldwide, requiring the advancement of early detection methods to improve prognosis and treatment outcomes. In this sense, mammography is regarded as the gold standard in breast cancer screening and early detection. However, in a scenario where extensive analysis is required, a large set of mammograms conducted by radiologists may carry out false negative or false positive diagnoses. Therefore, artificial intelligence has emerged in recent years as a method for enhancing timing in breast cancer diagnosis. Nonetheless, preprocessing stages are required to prepare the mammography dataset to enhance learning models to correctly identify breast anomalies. In this paper, we introduce a novel method employing convolutional neural networks (CNNs) to segment the pectoral muscle in 1288 mediolateral oblique mammograms (MLOs), thereby addressing class imbalance and overfitting between classes, and dataset augmentation based on translation, rotation, and scale transformation. The effectiveness of the model was assessed through a confusion matrix and performance metrics, highlighting an average Dice coefficient of 0.98 and a Jaccard index of 0.96. The outcomes demonstrate the model capability to accurately identify three classes: pectoral muscle, breast, and background. This study emphasizes the importance of tackling class imbalance problems and augmenting data for the training of models for reliable early breast cancer detection.

Generation of micrograph-annotation pairs for steel microstructure recognition using the hybrid deep generative model in the case of an extremely small and imbalanced dataset

Insufficient annotated samples coupled with class imbalance problem largely restrict the wide application of deep learning (DL)-based approach in microstructure recognition and quantification. In this work, we present a micrograph augmentation approach using the hybrid deep generative model to generate SEM image-annotation pairs for the establishment of a large-scale and well-balanced augmentation dataset. In this method, a generator is established to produce the desired annotations and then a translator is trained to translate these synthetic annotations into high-quality SEM images. The proposed method is successfully applied to an extremely small and imbalanced additively manufactured (AM) steel dataset containing only one SEM image-annotation pair with a very low martensite/austenite (MA) fraction, to significantly augment the initial dataset and achieve a more balanced distribution of phase fraction. The effectiveness of the present method is well demonstrated by the fact that the extensibility of microstructure recognition model to unseen micrographs is improved through the utilization of synthetic data. Furthermore, the impact of synthetic data proportion on the model's performance and the underlying reasons for synthetic data to improve the extensibility of trained models are also discussed in detail.