Recent Advancements and Future Prospects in Active Deep Learning for Medical Image Segmentation and Classification

IEEE Access, Год журнала: 2023, Номер 11, С. 113623 - 113652

Опубликована: Янв. 1, 2023

Medical images are helpful for the diagnosis, treatment, and evaluation of diseases. Precise medical image segmentation improves diagnosis decision-making, aiding intelligent services better disease management recovery. Due to unique nature images, algorithms based on deep learning face problems such as sample imbalance, edge blur, false positives, negatives. In view these problems, researchers primarily improve network structure but rarely from unstructured aspect. The paper tackles challenges, accentuating limitations convolutional neural network-based methods proposing solutions reduce annotation costs, particularly in complex introduces improvement strategies solve Additionally, article latest learning-based applications analysis, covering segmentation, acquisition, enhancement, registration, classification. Moreover, provides an overview four cutting-edge models, namely (CNN), belief (DBN), stacked autoencoder (SAE), recurrent (RNN). study selection involved searching benchmark academic databases, collecting relevant literature appropriate indicator emphasizing DL-based classification approaches, evaluating performance metrics. research highlights clinicians' scholars' obstacles developing efficient accurate malignancy prognostic framework state-of-the-art deep-learning algorithms. Furthermore, future perspectives explored overcome challenges advance field analysis.

Язык: Английский

---

Medical long-tailed learning for imbalanced data: Bibliometric analysis

Computer Methods and Programs in Biomedicine, Год журнала: 2024, Номер 247, С. 108106 - 108106

Опубликована: Фев. 29, 2024

Язык: Английский

---

Enhancing Prognosis Accuracy for Ischemic Cardiovascular Disease Using K Nearest Neighbor Algorithm: A Robust Approach

IEEE Access, Год журнала: 2023, Номер 11, С. 97879 - 97895

Опубликована: Янв. 1, 2023

Ischemic Cardiovascular diseases are one of the deadliest in world. However, mortality rate can be significantly reduced if we detect disease precisely and effectively. Machine Learning (ML) models offer substantial assistance to individuals requiring early treatment detection realm cardiovascular health. In response this critical need, study developed a robust system predict ischemic accurately using ML-based algorithms. The dataset obtained from Kaggle encompasses comprehensive collection over 918 observations, encompassing 12 essential features crucial for predicting disease. contrast, much-existing research relies primarily on datasets comprising only 303 instances UCI repository. Six algorithms, including K Nearest Neighbors (KNN), Random Forest (RF), Logistic Regression (LR), Support Vector (SVM), Gaussian Naïve Bayes (GNB), Decision Trees (DT), trained heart data. effectiveness proposed methodologies is meticulously evaluated benchmarked against cutting-edge techniques, employing range performance criteria. empirical findings manifest that KNN classifier produced optimized results with 91.8% accuracy, 91.4% recall, 91.9% F1 score, 92.5% precision, AUC 90.27%.

Язык: Английский

---

Deep Transfer Learning-Based Automated Diabetic Retinopathy Detection Using Retinal Fundus Images in Remote Areas

International Journal of Computational Intelligence Systems, Год журнала: 2024, Номер 17(1)

Опубликована: Май 29, 2024

Abstract Diabetic retinopathy (DR) significantly burdens ophthalmic healthcare due to its wide prevalence and high diagnostic costs. Especially in remote areas with limited medical access, undetected DR cases are on the rise. Our study introduces an advanced deep transfer learning-based system for real-time detection using fundus cameras address this. This research aims develop efficient timely assistance patients, empowering them manage their health better. The proposed leverages imaging collect retinal images, which then transmitted processing unit effective disease severity classification. Comprehensive reports guide subsequent actions based identified stage. achieves by utilizing learning algorithms, specifically VGGNet. system’s performance is rigorously evaluated, comparing classification accuracy previous outcomes. experimental results demonstrate robustness of system, achieving impressive 97.6% during phase, surpassing existing approaches. Implementing automated has transformed dynamics, enabling early, cost-effective diagnosis millions. also streamlines patient prioritization, facilitating interventions early-stage cases.

Язык: Английский

---

A Unified Approach Addressing Class Imbalance in Magnetic Resonance Image for Deep Learning Models

IEEE Access, Год журнала: 2024, Номер 12, С. 27368 - 27384

Опубликована: Янв. 1, 2024

Medical image datasets, particularly those comprising Magnetic Resonance (MR) images, are essential for accurate diagnosis and treatment planning. However, these datasets often suffer from class imbalance, where certain classes of abnormalities have unequal representation. Models trained on imbalanced can be biased towards the prominent class, leading to misclassification. Addressing imbalance problems is crucial developing robust deep-learning MR analysis models. This research focuses problem in proposes a novel approach enhance deep learning We introduced unified equipped with selective attention mechanism, loss function, progressive resizing. The strategy identifies regions within underlying find feature maps, retaining only relevant activations minority class. Fine-tuning multiple hyperparameters was achieved using function that plays vital role enhancing overwhelming error performance accuracy common classes. To address imbalances phenomenon, we incorporate resizing dynamically adjust input size as model trains. dynamic nature helps handle improve overall performance. evaluates proposed approach's effectiveness by embedding it into five state-of-the-art CNN models: UNet, FCN, RCNN, SegNet, Deeplab-V3. For experimental purposes, selected diverse BUS2017, MICCAI 2015 head neck, ATLAS, BRATS 2015, Digital Database Thyroid Image (DDTI), evaluate against techniques. assessment reveals improved across all metrics different imaging datasets. DeepLab-V3 demonstrated best performance, achieving IoU, DSC, Precision, Recall scores 0.893, 0.953, 0.943, 0.944, respectively, BUS dataset. These indicate an improvement 5% 6% 4% precision, approximately recall compared baseline. most significant increases were observed ATLAS LiTS 2017 7% increase IoU DSC over baseline (DSC = 0.628, 0.695) dataset, 9%

Язык: Английский

---

Pediatric Pneumonia Recognition Using an Improved DenseNet201 Model with Multi-Scale Convolutions and Mish Activation Function

Algorithms, Год журнала: 2025, Номер 18(2), С. 98 - 98

Опубликована: Фев. 10, 2025

Pediatric pneumonia remains a significant global health issue, particularly in low- and middle-income countries, where it contributes substantially to mortality children under five. This study introduces deep learning model for pediatric diagnosis from chest X-rays that surpasses the performance of state-of-the-art methods reported recent literature. Using DenseNet201 architecture with Mish activation function multi-scale convolutions, was trained on dataset 5856 X-ray images, achieving high performance: 0.9642 accuracy, 0.9580 precision, 0.9506 sensitivity, 0.9542 F1 score, 0.9507 specificity. These results demonstrate advancement diagnostic precision efficiency within this domain. By highest accuracy score compared other work using same dataset, our approach offers tangible improvement resource-constrained environments access specialists sophisticated equipment is limited. While need high-quality datasets adequate computational resources general consideration applications, model’s demonstrably superior establishes new benchmark delivery more timely precise diagnoses, potential significantly enhance patient outcomes.

Язык: Английский

---

Addressing Class Imbalance Problem in Health Data Classification: Practical Application From an Oversampling Viewpoint

Applied Computational Intelligence and Soft Computing, Год журнала: 2025, Номер 2025(1)

Опубликована: Янв. 1, 2025

While analyzing health data is important for improving outcomes, class imbalance in datasets poses major challenges to machine learning classification models. This work, therefore, considers the problem stroke prediction using models such as K‐nearest neighbors, support vector machine, logistic regression, random forest, and decision tree. work balances dataset, thereby enhancing model performance, through various oversampling strategies: (RO), ADASYN, SMOTE, SMOTE–Tomek. Compared results of imbalanced all applied techniques enhanced correct events by ML model. Among these, RO–SVM with RBF kernel was best terms sensitivity, specificity, G‐mean, F1‐score, accuracy values, offering highest respective values 89.87%, 94.91%, 92.36%, 89.64%, 89.87%. After applying techniques, classifications were good enough classify status, especially minority class. study has highlighted importance issues datasets. Precise detection instances classes can be considerably employing implementation hybrid strategies effectively solve issues, which, turn, will help improve healthcare outcomes. Further research integrating more advanced deep into other imbalances encouraged further validate or refine approaches, effective handling substantially promote predictive performance analysis healthcare.

Язык: Английский

---

Synergistic transfer learning and adversarial networks for breast cancer diagnosis: benign vs. invasive classification

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Март 3, 2025

Current breast cancer diagnosis methods often face limitations such as high cost, time consumption, and inter-observer variability. To address these challenges, this research proposes a novel deep learning framework that leverages generative adversarial networks (GANs) for data augmentation transfer to enhance classification using convolutional neural (CNNs). The uses two-stage approach. First, conditional Wasserstein GAN (cWGAN) generates synthetic images based on clinical data, enhancing training stability enabling targeted feature incorporation. Second, traditional techniques (e.g., rotation, flipping, cropping) are applied both original images. A multi-scale technique is also employed, integrating three pre-trained CNNs (DenseNet-201, NasNetMobile, ResNet-101) with enrichment scheme, allowing the model capture features at various scales. was evaluated BreakHis dataset, achieving an accuracy of 99.2% binary 98.5% multi-class classification, significantly outperforming existing methods. This offers more efficient, cost-effective, accurate approach diagnosis. Future work will focus generalizing datasets it into diagnostic workflows.

Язык: Английский

---

Impact of imbalanced features on large datasets

Frontiers in Big Data, Год журнала: 2025, Номер 8

Опубликована: Март 13, 2025

The exponential growth of image and video data motivates the need for practical real-time content-based searching algorithms. Features play a vital role in identifying objects within images. However, feature-based classification faces challenge due to uneven class instance distribution. Ideally, each should have an equal number instances features ensure optimal classifier performance. real-world scenarios often exhibit imbalances. Thus, this article explores framework based on features, analyzing balanced imbalanced distributions. Through extensive experimentation, we examine impact imbalance performance, primarily large datasets. comprehensive evaluation shows that all models perform better with balancing compared using dataset, underscoring importance dataset model accuracy. Distributed Gaussian (D-GA) Poisson (D-PO) are found be most effective techniques, especially improving Random Forest (RF) SVM models. deep learning experiments also show improvement as such.

Язык: Английский

---

Dual examiner consistency learning with dynamic receptive fields and class-balance refinement for Barely-supervised brain tumor segmentation

Displays, Год журнала: 2025, Номер unknown, С. 103054 - 103054

Опубликована: Апрель 1, 2025

Язык: Английский

---