An integrated convolutional neural network with attention guidance for improved performance of medical image classification

Neural Computing and Applications, Journal Year: 2023, Volume and Issue: 36(4), P. 2067 - 2099

Published: Nov. 20, 2023

Language: Английский

---

Brain Tumor Categorization and Retrieval Using Deep Brain Incep Res Architecture Based Reinforcement Learning Network

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 130584 - 130600

Published: Jan. 1, 2023

The categorization and retrieval of brain tumors using Magnetic Resonance Imaging (MRI) is a difficult but necessary process for tumor diagnosis. In this study, reinforcement learning agent proposed that can interact with an environment includes images retrieve categorize the most comparable to unknown query image. This article proposes unique fuzzy Deep Learning (DL)-based Reinforcement (RL) strategy categorizing three types as well no tumors. Brain Incep Res Architecture 2.0 based Network (DBIRA2.0-RLN), Convolutional Neural (CNN)-based technique, benefits from novel architecture in which descriptors are established inception block effective skip-connection mapping arrangement. To improve efficiency DBIRA2.0-RLN, improved samples created by training testing system logic-based technique. lower dimension descriptor vector image retrieval, obtained DBIRA2.0 binary coded Multilinear Principal Component Analysis. produces preserves several layers, then used sequentially numerous units construct final retrieval. method's output tested dataset, accuracy rates meningioma tumor, glioma pituitary 97.1%, 98.7%, 94.3%, 100% respectively, indicating approach outperforms other approaches literature.

Language: Английский

---

Automated multi-class MRI brain tumor classification and segmentation using deformable attention and saliency mapping

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: March 8, 2025

In the diagnosis and treatment of brain tumors, automatic classification segmentation medical images play a pivotal role. Early detection facilitates timely intervention, significantly improving patient survival rates. This study introduces novel method for automated aiming to enhance both diagnostic accuracy efficiency. Magnetic Resonance (MR) imaging remains gold standard in clinical tumor diagnostics; however, it is time-intensive labor-intensive process. Consequently, integration detection, localization, methods not only desirable but essential. this research, we present framework that enables post-classification feature extraction, allowing first-time multiple types. To improve characterization, applied data augmentation techniques MR developed hierarchical multiscale deformable attention module (MS-DAM). model effectively captures irregular complex patterns, enhancing performance. Following classification, comprehensive process was conducted across large dataset, reinforcing model's role as decision support system. Utilizing Kaggle dataset containing 14 different types with highly similar morphologic structures, validated proposed efficacy. Compared existing multi-scale channel modules, MS-DAM achieved superior accuracy, exceeding 96.5%. presents promising approach tumors imaging, offering significant advancements clinics paving way more efficient, accurate, scalable methodologies.

Language: Английский

---

Explainable cost-sensitive deep neural networks for brain tumor detection from brain MRI images considering data imbalance

Multimedia Tools and Applications, Journal Year: 2025, Volume and Issue: unknown

Published: April 30, 2025

Language: Английский

---

Fusion‐Brain‐Net: A Novel Deep Fusion Model for Brain Tumor Classification

Brain and Behavior, Journal Year: 2025, Volume and Issue: 15(5)

Published: May 1, 2025

ABSTRACT Problem Brain tumors are among the most prevalent and lethal diseases. Early diagnosis precise treatment crucial. However, manual classification of brain is a laborious complex task. Aim This study aimed to develop fusion model address certain limitations previous works, such as covering diverse image modalities in various datasets. Method We presented hybrid transfer learning model, Fusion‐Brain‐Net, at automatic tumor classification. The proposed method included four stages: preprocessing data augmentation, deep feature extractions, fine‐tuning, Integrating pre‐trained CNN models, VGG16, ResNet50, MobileNetV2, enhanced comprehensive extraction while mitigating overfitting issues, improving model's performance. Results was rigorously tested verified on public datasets: Br35H, Figshare, Nickparvar, Sartaj. It achieved remarkable accuracy rates 99.66%, 97.56%, 97.08%, 93.74%, respectively. Conclusion numerical results highlight that should be further investigated for potential use computer‐aided diagnoses improve clinical decision‐making.

Language: Английский

---

A brain tumour classification on the magnetic resonance images using convolutional neural network based privacy‐preserving federated learning

International Journal of Imaging Systems and Technology, Journal Year: 2024, Volume and Issue: 34(1)

Published: Jan. 1, 2024

Abstract The healthcare industry has found it challenging to build a powerful global classification model due the scarcity and diversity of medical data. leading cause is privacy, which restricts data sharing among providers. Federated learning (FL) can contribute developing models by protecting privacy. This study tested various federated techniques in peer‐to‐peer setting classify brain Magnetic Resonance Images (MRI). authors propose aggregation strategies for FL, including Averaging (FedAvg), Quantum FL with FedAVG (QFedAvg) Fault Tolerant FedAvg (Ft‐FedAvg) differential privacy (Dp‐FedAvg). In each approach, custom Convolutional Neural Network (CNN) applied compute locally run nodes different parts same MRI dataset 10, 20 30 training test rounds. A central server CNN‐based three clients are included FL‐based tumour exchange combine weights on server, sent from local devices server. superiority performance proposed demonstrated comparing traditional methods metrics. Experimental results show that using approaches, FedAVg showed best 85.55% 84.60% success 10 rounds, respectively, while Ft‐FedAvg 85.80% rounds set. Statistical obtained approaches have superior regard accuracy robustness comparison others.

Language: Английский

---

A deep neural network-based method to predict J-integral for surface cracked plates under biaxial loading

Engineering Fracture Mechanics, Journal Year: 2024, Volume and Issue: 302, P. 110062 - 110062

Published: April 6, 2024

Language: Английский

---

Evaluating the Potential of Wavelet Pooling on Improving the Data Efficiency of Light-Weight CNNs

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 51199 - 51213

Published: Jan. 1, 2023

Wavelet pooling (WP) in neural network architectures has recently demonstrated more discriminative power than traditional methods.This is mainly because the latter suffer from spatial information loss while wavelet harnesses of spectral information.However, potential WP increasing data efficiency and extent this have not been investigated yet.Data refers to volume training required attain a certain performance level during inference, e.g., recognition accuracy.In research, we are concerned with evaluating light-weight architectures-MobileNets.Across wide variety seven datasets/applications including object (CIFAR-10, STL-10, CINIC-10, Intel Image Classification datasets) diagnostic imaging (colon diseases, brain tumors, malaria cell images datasets), considering classification accuracy as metric, show that achieves an average saving exceeds 30% compared techniques.For other measures, namely, precision, recall, F1-score, report for datasets 22% datasets.By focusing on architecture, research further emphasizes significance testing resources-challenged settings such applications edge computing green deep learning.

Language: Английский

---

Deeply supervised network for airborne LiDAR tree classification incorporating dual attention mechanisms

GIScience & Remote Sensing, Journal Year: 2024, Volume and Issue: 61(1)

Published: Jan. 18, 2024

Accurately identifying tree species is crucial in digital forestry. Several airborne LiDAR-based classification frameworks have been proposed to facilitate work this area, and they achieved impressive results. These models range from the of characterization parameters based on feature engineering extraction end-to-end deep learning. However, practical applications, loud noises a single sample at varying vertical heights can cause misjudgment between intraspecific samples, thereby limiting accuracy. This may be exacerbated by scanning conditions geographic environment. To address challenge, deeply supervised network (DSTCN) designed article, which introduced height-intensity dual attention mechanism deliver improved performance. DSTCN takes histogram descriptors each slice as input vector considers features combination with its height intensity information, utilizing different information gains more effectively, removing accuracy limitations imposed noise heights. Experimental results seven mixed forest Baden-Württemberg, southwestern Germany indicate that (MAF = 0.94, OA Kappa 0.93, FISD 0.02) outperforms two commonly used methods, Point Net++ 0.88, 0.86, 0.08) BP Net 0.87, 0.85, 0.06) respectively, terms accuracy, stability, robustness. method integrates achieve precise balanced outcomes species. The simplified design enables efficient forestry decision-making presents innovative ideas for employing LiDAR technology identification large-scale multi-layer stands.

Language: Английский

---

MRI-based brain tumor ensemble classification using two stage score level fusion and CNN models

Egyptian Informatics Journal, Journal Year: 2024, Volume and Issue: 28, P. 100565 - 100565

Published: Nov. 7, 2024

Language: Английский

---