A survey on deep learning tools dealing with data scarcity: definitions, challenges, solutions, tips, and applications

Journal Of Big Data, Год журнала: 2023, Номер 10(1)

Опубликована: Апрель 14, 2023

Abstract Data scarcity is a major challenge when training deep learning (DL) models. DL demands large amount of data to achieve exceptional performance. Unfortunately, many applications have small or inadequate train frameworks. Usually, manual labeling needed provide labeled data, which typically involves human annotators with vast background knowledge. This annotation process costly, time-consuming, and error-prone. every framework fed by significant automatically learn representations. Ultimately, larger would generate better model its performance also application dependent. issue the main barrier for dismissing use DL. Having sufficient first step toward any successful trustworthy application. paper presents holistic survey on state-of-the-art techniques deal models overcome three challenges including small, imbalanced datasets, lack generalization. starts listing techniques. Next, types architectures are introduced. After that, solutions address listed, such as Transfer Learning (TL), Self-Supervised (SSL), Generative Adversarial Networks (GANs), Model Architecture (MA), Physics-Informed Neural Network (PINN), Deep Synthetic Minority Oversampling Technique (DeepSMOTE). Then, these were followed some related tips about acquisition prior purposes, well recommendations ensuring trustworthiness dataset. The ends list that suffer from scarcity, several alternatives proposed in order more each Electromagnetic Imaging (EMI), Civil Structural Health Monitoring, Medical imaging, Meteorology, Wireless Communications, Fluid Mechanics, Microelectromechanical system, Cybersecurity. To best authors’ knowledge, this review offers comprehensive overview strategies tackle

Язык: Английский

---

LViT: Language Meets Vision Transformer in Medical Image Segmentation

IEEE Transactions on Medical Imaging, Год журнала: 2023, Номер 43(1), С. 96 - 107

Опубликована: Июль 3, 2023

Deep learning has been widely used in medical image segmentation and other aspects. However, the performance of existing models limited by challenge obtaining sufficient high-quality labeled data due to prohibitive annotation cost. To alleviate this limitation, we propose a new text-augmented model LViT (Language meets Vision Transformer). In our model, text is incorporated compensate for quality deficiency data. addition, information can guide generate pseudo labels improved semi-supervised learning. We also an Exponential Pseudo label Iteration mechanism (EPI) help Pixel-Level Attention Module (PLAM) preserve local features setting. LV (Language-Vision) loss designed supervise training unlabeled images using directly. For evaluation, construct three multimodal datasets (image + text) containing X-rays CT images. Experimental results show that proposed superior both fully-supervised The code are available at https://github.com/HUANGLIZI/LViT .

Язык: Английский

---

Deep Learning for Medical Image-Based Cancer Diagnosis

Cancers, Год журнала: 2023, Номер 15(14), С. 3608 - 3608

Опубликована: Июль 13, 2023

(1) Background: The application of deep learning technology to realize cancer diagnosis based on medical images is one the research hotspots in field artificial intelligence and computer vision. Due rapid development methods, requires very high accuracy timeliness as well inherent particularity complexity imaging. A comprehensive review relevant studies necessary help readers better understand current status ideas. (2) Methods: Five radiological images, including X-ray, ultrasound (US), computed tomography (CT), magnetic resonance imaging (MRI), positron emission (PET), histopathological are reviewed this paper. basic architecture classical pretrained models comprehensively reviewed. In particular, advanced neural networks emerging recent years, transfer learning, ensemble (EL), graph network, vision transformer (ViT), introduced. overfitting prevention methods summarized: batch normalization, dropout, weight initialization, data augmentation. image-based analysis sorted out. (3) Results: Deep has achieved great success diagnosis, showing good results image classification, reconstruction, detection, segmentation, registration, synthesis. However, lack high-quality labeled datasets limits role faces challenges rare multi-modal fusion, model explainability, generalization. (4) Conclusions: There a need for more public standard databases cancer. pre-training potential be improved, special attention should paid multimodal fusion supervised paradigm. Technologies such ViT, few-shot will bring surprises images.

Язык: Английский

---

A Transfer Learning Approach with a Convolutional Neural Network for the Classification of Lung Carcinoma

Healthcare, Год журнала: 2022, Номер 10(6), С. 1058 - 1058

Опубликована: Июнь 8, 2022

Lung cancer is among the most hazardous types of in humans. The correct diagnosis pathogenic lung disease critical for medication. Traditionally, determining pathological form involves an expensive and time-consuming process investigation. a leading cause mortality worldwide, with tissue nodules being prevalent way doctors to identify it. proposed model based on robust deep-learning-based detection recognition. This study uses deep neural network as extraction features approach computer-aided diagnosing (CAD) system assist detecting illnesses at high definition. categorized into three phases: first, data augmentation performed, classification then performed using pretrained CNN model, lastly, localization completed. amount obtained medical image assessment occasionally inadequate train learning network. We classifier technique known transfer (TL) solve issue introduced process. methodology offers non-invasive diagnostic tool use clinical that effective. has lower number parameters are much smaller compared state-of-the-art models. also examined desired dataset's robustness depending its size. standard performance metrics used assess effectiveness architecture. In this dataset, all TL techniques perform well, VGG 16, 19, Xception 20 epoch structure compared. Preprocessing functions wonderful bridge build dependable eventually helps forecast future scenarios by including interface faster phase any model. At 20th epoch, accuracy 98.83 percent, 98.05 97.4 percent.

Язык: Английский

---

Video surveillance using deep transfer learning and deep domain adaptation: Towards better generalization

Engineering Applications of Artificial Intelligence, Год журнала: 2022, Номер 119, С. 105698 - 105698

Опубликована: Дек. 16, 2022

Recently, developing automated video surveillance systems (VSSs) has become crucial to ensure the security and safety of population, especially during events involving large crowds, such as sporting events. While artificial intelligence (AI) smooths path computers think like humans, machine learning (ML) deep (DL) pave way more, even by adding training components. DL algorithms require data labeling high-performance effectively analyze understand recorded from fixed or mobile cameras installed in indoor outdoor environments. However, they might not perform expected, take much time training, have enough input generalize well. To that end, transfer (DTL) domain adaptation (DDA) recently been proposed promising solutions alleviate these issues. Typically, can (i) ease process, (ii) improve generalizability ML models, (iii) overcome scarcity problems transferring knowledge one another task another. Although increasing number articles develop DTL- DDA-based VSSs, a thorough review summarizes criticizes state-of-the-art is still missing. this paper introduces, best authors' knowledge, first overview existing shed light on their benefits, discuss challenges, highlight future perspectives.

Язык: Английский

---

Alexnet architecture variations with transfer learning for classification of wound images

Engineering Science and Technology an International Journal, Год журнала: 2023, Номер 45, С. 101490 - 101490

Опубликована: Июль 28, 2023

In medical world, wound care and follow-up is one of the issues that are gaining importance to work on day by day. Accurate early recognition wounds can reduce treatment costs. field computer vision, deep learning architectures have received great attention recently. The achievements existing pre-trained for describing (classifying) data belonging many image sets in real world primarily addressed. However, increase success these a certain area, some improvements enhancements be made architecture. this paper, classification pressure diabetic images was performed with high accuracy. six different new AlexNet architecture variations (3Conv_Softmax, 3Conv_SVM, 4Conv_Softmax, 4Conv_SVM, 6Conv_Softmax, 6Conv_SVM) were created number implementations Convolution, Pooling, Rectified Linear Activation (ReLU) layers. Classification performances proposed models investigated using Softmax classifier SVM separately. A original Wound Image Database performance measures. According experimental results obtained Database, model 6 Convolution layers (6Conv_SVM) most successful method among methods 98.85% accuracy, 98.86% sensitivity, 99.42% specificity. 6Conv_SVM also tested public medetec dataset, 95.33% 97.66% specificity values obtained. provides compared other state-of-the-art literature. showed used relevant departments good tasks such as examining classifying following up process.

Язык: Английский

---

A systematic review of deep learning-based cervical cytology screening: from cell identification to whole slide image analysis

Artificial Intelligence Review, Год журнала: 2023, Номер 56(S2), С. 2687 - 2758

Опубликована: Окт. 5, 2023

Abstract Cervical cancer is one of the most common cancers in daily life. Early detection and diagnosis can effectively help facilitate subsequent clinical treatment management. With growing advancement artificial intelligence (AI) deep learning (DL) techniques, an increasing number computer-aided (CAD) methods based on have been applied cervical cytology screening. In this paper, we survey more than 80 publications since 2016 to provide a systematic comprehensive review DL-based First, concise summary medical biological knowledge pertaining cytology, hold firm belief that biomedical understanding significantly contribute development CAD systems. Then, collect wide range public datasets. Besides, image analysis approaches applications including cell identification, abnormal or area detection, region segmentation whole slide are summarized. Finally, discuss present obstacles promising directions for future research automated

Язык: Английский

---

VER-Net: a hybrid transfer learning model for lung cancer detection using CT scan images

BMC Medical Imaging, Год журнала: 2024, Номер 24(1)

Опубликована: Май 24, 2024

Abstract Background Lung cancer is the second most common worldwide, with over two million new cases per year. Early identification would allow healthcare practitioners to handle it more effectively. The advancement of computer-aided detection systems significantly impacted clinical analysis and decision-making on human disease. Towards this, machine learning deep techniques are successfully being applied. Due several advantages, transfer has become popular for disease based image data. Methods In this work, we build a novel model (VER-Net) by stacking three different models detect lung using CT scan images. trained map images four classes. Various measures, such as preprocessing, data augmentation, hyperparameter tuning, taken improve efficacy VER-Net. All evaluated multiclass classifications chest Results experimental results confirm that VER-Net outperformed other eight compared with. scored 91%, 92%, 91.3% when tested accuracy, precision, recall, F1-score, respectively. Compared state-of-the-art, better accuracy. Conclusion not only effectively used but may also be useful diseases which available.

Язык: Английский

---

Generalizable disease detection using model ensemble on chest X-ray images

Scientific Reports, Год журнала: 2024, Номер 14(1)

Опубликована: Март 11, 2024

Abstract In the realm of healthcare, demand for swift and precise diagnostic tools has been steadily increasing. This study delves into a comprehensive performance analysis three pre-trained convolutional neural network (CNN) architectures: ResNet50, DenseNet121, Inception-ResNet-v2. To ensure broad applicability our approach, we curated large-scale dataset comprising diverse collection chest X-ray images, that included both positive negative cases COVID-19. The models’ was evaluated using separate datasets internal validation (from same source as training images) external different sources). Our examination uncovered significant drop in efficacy, registering 10.66% reduction 36.33% decline 19.55% decrease Inception-ResNet-v2 terms accuracy. Best results were obtained with DenseNet121 achieving highest accuracy at 96.71% attaining 76.70% validation. Furthermore, introduced model ensemble approach aimed improving when making inferences on images from sources beyond their data. proposed method uses uncertainty-based weighting by calculating entropy order to assign appropriate weights outputs each network. showcase effectiveness enhancing up 97.38% 81.18% validation, while maintaining balanced ability detect cases.

Язык: Английский

---

A Review of Predictive Analytics Models in the Oil and Gas Industries

Sensors, Год журнала: 2024, Номер 24(12), С. 4013 - 4013

Опубликована: Июнь 20, 2024

Enhancing the management and monitoring of oil gas processes demands development precise predictive analytic techniques. Over past two years, its prediction have advanced significantly using conventional modern machine learning Several review articles detail developments in maintenance technical non-technical aspects influencing uptake big data. The absence references for techniques impacts effective optimization analytics sectors. This paper offers readers thorough information on latest methods utilized this industry’s analytical modeling. covers different forms used modeling from 2021 to 2023 (91 articles). It provides an overview details papers that were reviewed, describing model’s categories, data’s temporality, field, name, dataset’s type, (classification, clustering, or prediction), models’ input output parameters, performance metrics, optimal model, benefits drawbacks. In addition, suggestions future research directions provide insights into potential applications associated knowledge. can serve as a guide enhance effectiveness models industries.

Язык: Английский

---