Evaluating Deep Learning Architectures for Breast Tumor Classification and Ultrasound Image Detection Using Transfer Learning

Big Data and Cognitive Computing, Journal Year: 2025, Volume and Issue: 9(5), P. 111 - 111

Published: April 23, 2025

The intersection of medical image classification and deep learning has garnered increasing research interest, particularly in the context breast tumor detection using ultrasound images. Prior studies have predominantly focused on classification, segmentation, feature extraction, often assuming that input images, whether sourced from healthcare professionals or individuals, are valid relevant for analysis. To address this, we propose an initial binary filter to distinguish between irrelevant ensuring only meaningful data proceeds subsequent However, primary focus this study lies investigating performance a hierarchical two-tier architecture compared traditional flat three-class model, by employing well-established images dataset. Specifically, explore sequentially breaking down problem into classifications, first identifying normal versus tumorous tissue then distinguishing benign malignant tumors, yields better accuracy robustness than directly classifying all three categories single step. Using range evaluation metrics, demonstrates notable advantages certain critical aspects model performance. findings provide valuable guidance selecting optimal final facilitating its seamless integration web application deployment. These insights further anticipated advance future algorithm development broaden potential applicability across diverse fields.

Language: Английский

---

A prior knowledge-guided distributionally robust optimization-based adversarial training strategy for medical image classification

Information Sciences, Journal Year: 2024, Volume and Issue: 673, P. 120705 - 120705

Published: May 7, 2024

Language: Английский

---

Automated Breast Cancer Detection: A Review

Communications in computer and information science, Journal Year: 2025, Volume and Issue: unknown, P. 297 - 310

Published: Jan. 1, 2025

Language: Английский

---

Generating Pseudo-Subtracted Image in Dual-Energy Contrast-Enhanced Spectral Mammography Using Transfer Learning

Research Square (Research Square), Journal Year: 2025, Volume and Issue: unknown

Published: March 26, 2025

Background: Dual-energy contrast-enhanced spectral mammography (CESM) enhances breast cancer detection but increases radiation exposure, especially for high-risk patients like BRCA1 mutation carriers. Additionally, the dual-energy acquisition process can be time-consuming. This study uses deep learning to convert low-energy images into subtracted images, reducing and contrast-related risks, while also addressing time consumption challenge of traditional CESM procedure. Methods: The utilized Categorized Digital Database Low-energy Subtracted Contrast-Enhanced Spectral Mammography Images (CDD-CESM), which contains 7600 image pairs after augmentation. dataset was divided 70% training 30% testing. CycleGAN's performance evaluated compared against U-Net, Pix2Pix, ResNet18. Key metrics comparison included Structural Similarity Index Peak Signal-to-Noise Ratio. models were tested their ability generate high-quality without need paired data. Results: CycleGAN outperformed ResNet18 in generating pseudo-subtracted images. SSIM score 0.961, close that real indicates successfully preserves structural details. achieved this at a lower computational cost Conclusions: effectively generates from data, presenting viable alternative imaging. method has potential reduce additional imaging, minimize simplify imaging procedures. high highlights maintain strong similarities generated making it promising tool detecting lesions mammography.

Language: Английский

---

Boosting Skin Cancer Classification: A Multi-Scale Attention and Ensemble Approach with Vision Transformers

Sensors, Journal Year: 2025, Volume and Issue: 25(8), P. 2479 - 2479

Published: April 15, 2025

Skin cancer is a significant global health concern, with melanoma being the most dangerous form, responsible for majority of skin cancer-related deaths. Early detection critical, as it can drastically improve survival rates. While deep learning models have achieved impressive results in classification, there remain challenges accurately distinguishing between benign and malignant lesions. In this study, we introduce novel multi-scale attention-based performance booster inspired by Vision Transformer (ViT) architecture, which enhances accuracy both ViT convolutional neural network (CNN) models. By leveraging attention maps to identify discriminative regions within lesion images, our method improves models’ focus on diagnostically relevant areas. Additionally, employ ensemble techniques combine outputs several using voting. Our classifier, consisting EfficientNet models, classification 95.05% ISIC2018 dataset, outperforming individual The demonstrate effectiveness integrating methods classification.

Language: Английский

---

Deep learning in vibrational spectroscopy: Benefits, limitations, and recent progress

Journal of the Chinese Chemical Society, Journal Year: 2025, Volume and Issue: unknown

Published: May 9, 2025

Abstract Vibrational spectroscopy is a cornerstone in molecular analysis, offering detailed insights into chemical compositions and dynamics. Recent years have witnessed paradigm shift with the integration of deep learning, which excels automatically extracting intricate patterns from raw spectral data, bypassing traditional preprocessing steps. This synergy has significantly enhanced precision speed applications ranging material science to biomedical diagnostics. review comprehensively explores transformative impact learning on vibrational modeling, emphasizing its superiority over machine approaches. However, interplay between still presents significant challenges, including demand for massive labeled datasets, risk overfitting, particularly limited samples, inherently black‐box nature models. To address these this highlights recent breakthroughs that leverage unique two fields. For instance, transfer enables knowledge across domains, mitigating need extensive data. Generative adversarial networks synthetically expand datasets by capturing complex inherent spectra. Physics‐informed neural integrate spectroscopic principles directly model architectures, bridging gap physical data‐driven Additionally, enhancing interpretability through techniques like attention mechanisms saliency mapping critical trustworthy deployment, especially high‐stakes where domain‐specific can guide validate predictions. not only encapsulates advancements but also distills best practices development, experimental design tailored hyperparameter tuning robustness, validation protocols ensure reliability cheminformatics. provides an overview latest research past 2 offers future directions modeling face big data challenges.

Language: Английский

---

Improved Breast Cancer Classification Approach Using Hybrid Deep Learning Strategies for Tumor Segmentation

Sensing and Imaging, Journal Year: 2024, Volume and Issue: 25(1)

Published: June 1, 2024

Language: Английский

---

Ensemble of Deep Features for Breast Cancer Histopathological Image Classification

The Computer Journal, Journal Year: 2024, Volume and Issue: 67(6), P. 2126 - 2136

Published: Jan. 14, 2024

Abstract Analysis of histopathological images (HIs) is crucial for detecting breast cancer (BR). However, because they vary, it still very difficult to extract well-designed elements. Deep learning (DL) a recent development that used high-level features. DL techniques continue confront several problems, such as the need sufficient training data models, which reduces classification findings. In this study, an ensemble deep transfer convolutional neural network presented address problem. The pre-trained models (ResNet50 and MobileNet) are employed features by freezing front layer parameters while fine-tuning last layers. proposed framework, KNN, SVM, logistic regression networks base classifiers. majority vote product approaches integrate predictions each separate classifier. benchmark BreaKHis dataset, suggested model compared some current approaches. It demonstrates obtains considerable accuracy 97.72% multiclass test, achieves 99.2% binary task. model’s effectiveness in extracting useful BR demonstrated comparison with existing cutting-edge models.

Language: Английский

---

Enhancing Cancer Classification through a Hybrid Bio-Inspired Evolutionary Algorithm for Biomarker Gene Selection

Computers, materials & continua/Computers, materials & continua (Print), Journal Year: 2024, Volume and Issue: 79(1), P. 675 - 694

Published: Jan. 1, 2024

In this study, our aim is to address the problem of gene selection by proposing a hybrid bio-inspired evolutionary algorithm that combines Grey Wolf Optimization (GWO) with Harris Hawks (HHO) for feature selection.The motivation utilizing GWO and HHO stems from their nature demonstrated success in optimization problems.We leverage strengths these algorithms enhance effectiveness microarray-based cancer classification.We selected leave-one-out cross-validation (LOOCV) evaluate performance both two widely used classifiers, k-nearest neighbors (KNN) support vector machine (SVM), on high-dimensional microarray data.The proposed method extensively tested six publicly available datasets, comprehensive comparison recently published methods conducted.Our demonstrates its improving classification performance, Surpassing alternative approaches terms precision.The outcomes confirm capability substantially improve precision efficiency classification, thereby advancing development more efficient treatment strategies.The offers promising solution classification.It improves accuracy diagnosis treatment, superior compared other highlights potential applicability realworld tasks.By harnessing complementary search mechanisms HHO, we behavior identify informative genes relevant treatment.

Language: Английский

---

Enhanced cervical precancerous lesions detection and classification using Archimedes Optimization Algorithm with transfer learning

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: May 27, 2024

Abstract Cervical cancer (CC) ranks as the fourth most common form of affecting women, manifesting in cervix. CC is caused by Human papillomavirus (HPV) infection and eradicated vaccinating women from an early age. However, limited medical facilities present a significant challenge mid- or low-income countries. It can improve survivability rate be successfully treated if detected at earlier stages. Current technological improvements allow for cost-effective, more sensitive, rapid screening treatment measures CC. DL techniques are widely adopted automated detection architectures used to detect provide higher performance. This study offers design Enhanced Precancerous Lesions Detection Classification using Archimedes Optimization Algorithm with Transfer Learning (CPLDC-AOATL) algorithm. The CPLDC-AOATL algorithm aims diagnose cervical images. At preliminary stage, technique involves bilateral filtering (BF) eliminate noise input Besides, applies Inception-ResNetv2 model feature extraction process, use AOA chose hyperparameters. bidirectional long short-term memory (BiLSTM) process. experimental outcome emphasized superior accuracy 99.53% over other existing approaches under benchmark dataset.

Language: Английский

---