Ten deep learning techniques to address small data problems with remote sensing

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2023, Volume and Issue: 125, P. 103569 - 103569

Published: Nov. 18, 2023

Researchers and engineers have increasingly used Deep Learning (DL) for a variety of Remote Sensing (RS) tasks. However, data from local observations or via ground truth is often quite limited training DL models, especially when these models represent key socio-environmental problems, such as the monitoring extreme, destructive climate events, biodiversity, sudden changes in ecosystem states. Such cases, also known small pose significant methodological challenges. This review summarises challenges RS domain possibility using emerging techniques to overcome them. We show that problem common challenge across disciplines scales results poor model generalisability transferability. then introduce an overview ten promising techniques: transfer learning, self-supervised semi-supervised few-shot zero-shot active weakly supervised multitask process-aware ensemble learning; we include validation technique spatial k-fold cross validation. Our particular contribution was develop flowchart helps users select which use given by answering few questions. hope our article facilitate applications tackle societally important environmental problems with reference data.

Language: Английский

---

Deep Learning for Automated Detection and Localization of Traumatic Abdominal Solid Organ Injuries on CT Scans

Deleted Journal, Journal Year: 2024, Volume and Issue: 37(3), P. 1113 - 1123

Published: Feb. 16, 2024

Abstract Computed tomography (CT) is the most commonly used diagnostic modality for blunt abdominal trauma (BAT), significantly influencing management approaches. Deep learning models (DLMs) have shown great promise in enhancing various aspects of clinical practice. There limited literature available on use DLMs specifically image evaluation. In this study, we developed a DLM aimed at detecting solid organ injuries to assist medical professionals rapidly identifying life-threatening injuries. The study enrolled patients from single center who received CT scans between 2008 and 2017. Patients with spleen, liver, or kidney injury were categorized as group, while others considered negative cases. Only images acquired enrolled. A subset last year was designated test set, remaining utilized train validate detection models. performance each model assessed using metrics such area under receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value, value based best Youden index point. 1302 (87%) training tested them 194 (13%) scans. spleen demonstrated an accuracy 0.938 specificity 0.952. liver reported 0.820 0.847, respectively. showed 0.959 0.989. We that can automate by acceptable accuracy. It cannot replace role clinicians, but expect it be potential tool accelerate process therapeutic decisions care.

Language: Английский

---

Brain tumor segmentation using neuro-technology enabled intelligence-cascaded U-Net model

Frontiers in Computational Neuroscience, Journal Year: 2024, Volume and Issue: 18

Published: April 3, 2024

According to experts in neurology, brain tumours pose a serious risk human health. The clinical identification and treatment of rely heavily on accurate segmentation. varied sizes, forms, locations make automated segmentation formidable obstacle the field neuroscience. U-Net, with its computational intelligence concise design, has lately been go-to model for fixing medical picture issues. Problems restricted local receptive fields, lost spatial information, inadequate contextual information are still plaguing artificial intelligence. A convolutional neural network (CNN) Mel-spectrogram basis this cough recognition technique. First, we combine voice variety intricate settings improve audio data. After that, preprocess data sure length is consistent create out it. novel tumor (BTS), Intelligence Cascade U-Net (ICU-Net), proposed address these It built dynamic convolution uses non-local attention mechanism. In order reconstruct more detailed tumours, principal design two-stage cascade 3DU-Net. paper’s objective identify best learnable parameters that will maximize likelihood network’s ability gather long-distance dependencies AI, Expectation–Maximization applied lateral connections, enabling it leverage effectively. Lastly, enhance capture characteristics, convolutions adaptive capabilities used place standard convolutions. We compared our results those other typical methods ran extensive testing utilising publicly available BraTS 2019/2020 datasets. suggested method performs well tasks involving BTS, according experimental Dice scores core (TC), complete tumor, enhanced validation sets 0.897/0.903, 0.826/0.828, 0.781/0.786, respectively, indicating high performance BTS.

Language: Английский

---

MPEK: a multitask deep learning framework based on pretrained language models for enzymatic reaction kinetic parameters prediction

Briefings in Bioinformatics, Journal Year: 2024, Volume and Issue: 25(5)

Published: July 25, 2024

Enzymatic reaction kinetics are central in analyzing enzymatic mechanisms and target-enzyme optimization, thus biomanufacturing other industries. The enzyme turnover number (kcat) Michaelis constant (Km), key kinetic parameters for measuring catalytic efficiency, crucial the directed evolution of target enzymes. Experimental determination kcat Km is costly terms time, labor, cost. To consider intrinsic connection between further improve prediction performance, we propose a universal pretrained multitask deep learning model, MPEK, to predict these simultaneously while considering pH, temperature, organismal information. Through testing on same test datasets, MPEK demonstrated superior performance over previous models. Specifically, achieved Pearson coefficient 0.808 predicting kcat, improving ca. 14.6% 7.6% compared DLKcat UniKP models, it 0.777 Km, 34.9% 53.3% Kroll_model More importantly, was able reveal promiscuity sensitive slight changes mutant sequence. In addition, three case studies, shown that has potential assisted mining evolution. facilitate silico evaluation have established web server implementing this which can be accessed at http://mathtc.nscc-tj.cn/mpek.

Language: Английский

---

QGFormer: Queries-guided transformer for flexible medical image synthesis with domain missing

Expert Systems with Applications, Journal Year: 2024, Volume and Issue: 247, P. 123318 - 123318

Published: Jan. 23, 2024

Language: Английский

---

Survey of Transfer Learning Approaches in the Machine Learning of Digital Health Sensing Data

Journal of Personalized Medicine, Journal Year: 2023, Volume and Issue: 13(12), P. 1703 - 1703

Published: Dec. 12, 2023

Machine learning and digital health sensing data have led to numerous research achievements aimed at improving technology. However, using machine in poses challenges related availability, such as incomplete, unstructured, fragmented data, well issues privacy, security, format standardization. Furthermore, there is a risk of bias discrimination models. Thus, developing an accurate prediction model from scratch can be expensive complicated task that often requires extensive experiments complex computations. Transfer methods emerged feasible solution address these by transferring knowledge previously trained develop high-performance models for new task. This survey paper provides comprehensive study the effectiveness transfer applications enhance accuracy efficiency diagnoses prognoses, improve healthcare services. The first part this presents discusses most common technologies valuable resources applications, including learning. second meaning learning, clarifying categories types transfer. It also explains strategies, their role addressing models, specifically on data. These include feature extraction, fine-tuning, domain adaptation, multitask federated few-/single-/zero-shot highlights key features each method strategy, limitations applications. Overall, which aims inspire researchers gain approaches health, current strategies overcome limitations, apply them variety technologies.

Language: Английский

---

Enhancing Fetal Anomaly Detection in Ultrasonography Images: A Review of Machine Learning-Based Approaches

Biomimetics, Journal Year: 2023, Volume and Issue: 8(7), P. 519 - 519

Published: Nov. 2, 2023

Fetal development is a critical phase in prenatal care, demanding the timely identification of anomalies ultrasound images to safeguard well-being both unborn child and mother. Medical imaging has played pivotal role detecting fetal abnormalities malformations. However, despite significant advances technology, accurate irregularities continues pose considerable challenges, often necessitating substantial time expertise from medical professionals. In this review, we go through recent developments machine learning (ML) methods applied images. Specifically, focus on range ML algorithms employed context ultrasound, encompassing tasks such as image classification, object recognition, segmentation. We highlight how these innovative approaches can enhance ultrasound-based anomaly detection provide insights for future research clinical implementations. Furthermore, emphasize need further domain where investigations contribute more effective detection.

Language: Английский

---

Knowledge‐based deep learning system for classifying Alzheimer's disease for multi‐task learning

CAAI Transactions on Intelligence Technology, Journal Year: 2024, Volume and Issue: 9(4), P. 805 - 820

Published: Feb. 8, 2024

Abstract Deep learning has recently become a viable approach for classifying Alzheimer's disease (AD) in medical imaging. However, existing models struggle to efficiently extract features from images and may squander additional information resources illness classification. To address these issues, deep three‐dimensional convolutional neural network incorporating multi‐task attention mechanisms is proposed. An upgraded primary C3D utilised create rougher low‐level feature maps. It introduces new convolution block that focuses on the structural aspects of magnetic resonance imaging image another extracts weights unique certain pixel positions map multiplies them with output. Then, several fully connected layers are used achieve learning, generating three outputs, including classification task. The other two outputs employ backpropagation during training improve job. Experimental findings show authors’ proposed method outperforms current approaches AD, achieving enhanced accuracy indicators Neuroimaging Initiative dataset. authors demonstrate promise future studies.

Language: Английский

---

MPEK: a multi-task learning based on pre-trained language model for predicting enzymatic reaction kinetic parameters

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: Feb. 15, 2024

Abstract Enzymatic reaction kinetics are central in analyzing enzymatic mechanisms and target-enzyme optimization, thus biomanufacturing other industries. The enzyme turnover number ( k cat ) Michaelis constant K m ), key kinetic parameters for measuring catalytic efficiency crucial the directed evolution of target enzymes. Experimental determination is costly terms time, labor, cost. To consider intrinsic connection between further improve prediction performance , we propose a universal pre-trained multi-task deep learning model, MPEK, to predict these simultaneously while considering pH, temperature, organismal information. MPEK achieved superior predictive on whole test dataset. Using same dataset, outperformed state-of-the-art models. More importantly, was able reveal promiscuity sensitive slight changes mutant sequence. In addition, three case studies, it shown has potential assisted mining evolution. facilitate silico evaluation efficiency, have established web server implementing this model (http://mathtc.nscc-tj.cn/mpek).

Language: Английский

---

A dual-branch joint learning network for underwater object detection

Knowledge-Based Systems, Journal Year: 2024, Volume and Issue: 293, P. 111672 - 111672

Published: March 20, 2024

Language: Английский

---