Advancing breast cancer diagnosis: token vision transformers for faster and accurate classification of histopathology images

Visual Computing for Industry Biomedicine and Art, Journal Year: 2025, Volume and Issue: 8(1)

Published: Jan. 8, 2025

Abstract The vision transformer (ViT) architecture, with its attention mechanism based on multi-head layers, has been widely adopted in various computer-aided diagnosis tasks due to effectiveness processing medical image information. ViTs are notably recognized for their complex which requires high-performance GPUs or CPUs efficient model training and deployment real-world diagnostic devices. This renders them more intricate than convolutional neural networks (CNNs). difficulty is also challenging the context of histopathology analysis, where images both limited complex. In response these challenges, this study proposes a TokenMixer hybrid-architecture that combines strengths CNNs ViTs. hybrid architecture aims enhance feature extraction classification accuracy shorter time fewer parameters by minimizing number input patches employed during training, while incorporating tokenization using layers encoder process across all network fast accurate breast cancer tumor subtype classification. inspired ConvMixer TokenLearner models. First, dynamically generates spatial maps enabling from minimize used training. Second, extracts relevant regions selected patches, tokenizes improve extraction, trains tokenized an network. We evaluated BreakHis public dataset, comparing it ViT-based other state-of-the-art methods. Our approach achieved impressive results binary multi-classification subtypes magnification levels (40×, 100×, 200×, 400×). demonstrated accuracies 97.02% 93.29% multi-classification, decision times 391.71 1173.56 s, respectively. These highlight potential our deep ViT-CNN advancing histopathological images. source code accessible: https://github.com/abimouloud/TokenMixer .

Language: Английский

---

High precision banana variety identification using vision transformer based feature extraction and support vector machine

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

Published: March 26, 2025

Bananas, renowned for their delightful flavor, exceptional nutritional value, and digestibility, are among the most widely consumed fruits globally. The advent of advanced image processing, computer vision, deep learning (DL) techniques has revolutionized agricultural diagnostics, offering innovative automated solutions detecting classifying fruit varieties. Despite significant progress in DL, accurate classification banana varieties remains challenging, particularly due to difficulty identifying subtle features at early developmental stages. To address these challenges, this study presents a novel hybrid framework that integrates Vision Transformer (ViT) model global semantic feature representation with robust capabilities Support Vector Machines. proposed was rigorously evaluated on two datasets: four-class BananaImageBD six-class BananaSet. mitigate data imbalance issues, evaluation strategy employed, resulting remarkable accuracy rate (CAR) 99.86% $$\:\pm\:$$ 0.099 BananaSet 99.70% 0.17 BananaImageBD, surpassing traditional methods by margin 1.77%. ViT model, leveraging self-supervised semi-supervised mechanisms, demonstrated promise extracting nuanced critical applications. By combining cutting-edge machine classifiers, system establishes new benchmark precision reliability detection These findings underscore potential DL frameworks advancing diagnostics pave way future innovations domain.

Language: Английский

---

AI augmented edge and fog computing for Internet of Health Things (IoHT)

PeerJ Computer Science, Journal Year: 2025, Volume and Issue: 11, P. e2431 - e2431

Published: Jan. 30, 2025

Patients today seek a more advanced and personalized health-care system that keeps up with the pace of modern living. Cloud computing delivers resources over Internet enables deployment an infinite number applications to provide services many sectors. The primary limitation these cloud frameworks right now is their limited scalability, which results in inability meet needs. An edge/fog environment, paired current techniques, answer fulfill energy efficiency latency requirements for real-time collection analysis health data. Additionally, Things (IoT) revolution has been essential changing contemporary healthcare systems by integrating social, economic, technological perspectives. This requires transitioning from unadventurous adapted allow patients be identified, managed, evaluated easily. These techniques data sources integrated effectively assess patient status predict potential preventive actions. A subset Things, Health (IoHT) remote exchange physical processes like monitoring, treatment progress, observation, consultation. Previous surveys related mainly focused on architecture networking, left untouched important aspects smart optimal such as artificial intelligence, deep learning, technologies, includes 5G unified communication service (UCaaS). study aims examine future existing fog edge architectures methods have augmented intelligence (AI) use applications, well defining demands challenges incorporating technology IoHT, thereby helping professionals technicians identify relevant technologies required based need developing IoHT healthcare. Among crucial elements take into account framework are efficient resource management, low latency, strong security. review addresses several machine learning management IoT, where (ML) AI crucial. It noted how narrow band-IoT (NB-IoT) wider coverage Blockchain security, transforming IoHT. last part focuses posed services. provides prospective research suggestions enhancing order give improved quality life.

Language: Английский

---

Multi-classification of breast cancer histopathological image using enhanced shallow convolutional neural network

Journal of Engineering and Applied Science, Journal Year: 2025, Volume and Issue: 72(1)

Published: Feb. 14, 2025

Abstract Cancer of the breast popularly known as cancer (BC) is second and third utmost cause mortality among women in Nigeria globally, respectively. Biopsy histopathological images (BHI) have gained more attention for early clinical diagnosis BC. However, examination BC histology are subject to human error. Consequently, several computer-aided diagnoses (CAD) solutions been presented aid histopathologists with automated classification cancerous tumor cells on histological images. Deep convolutional neural networks (DCNN) utilized build a sizable portion cutting-edge proposed solutions. due architectural structure DCNN, which extracts features automatically along training processes coupled overlapping nucleic (BHI), existing suffer from high computational utilization, extensive time leading longer convergence times, reliance available high-end system resources adequate In this paper, an enhanced shallow network (ES-CNN) has multi-classification BHI, aimed improve performance reduce across eight types four magnifications BreakHis dataset. The research objectives were achieved three ways. First, we designed network’s architecture, guided by magnification patient dependencies. Secondly, implemented model based network, and, finally, two categories experiments conducted accuracy utilization. experimental results revealed that methods minimal utilization improved compared work. This reports 96%, 95%, 98%, 96% 400 × , 200 100 40 image magnifications,

Language: Английский

---

Enhanced Multi-Class Breast Cancer Classification from Whole-Slide Histopathology Images Using a Proposed Deep Learning Model

Diagnostics, Journal Year: 2025, Volume and Issue: 15(5), P. 582 - 582

Published: Feb. 27, 2025

Background/Objectives: Breast cancer is among the most frequently diagnosed cancers and leading cause of mortality worldwide. The accurate classification breast from histology photographs very important for diagnosis effective treatment planning. Methods: In this article, we propose a DenseNet121-based deep learning model detection multi-class classification. experiments were performed using whole-slide histopathology images collected BreakHis dataset. Results: proposed method attained state-of-the-art performance with 98.50% accuracy an AUC 0.98 binary classification, it obtained competitive results 92.50% 0.94. Conclusions: outperforms methods in distinguishing between benign malignant tumors as well classifying specific malignancy subtypes. This study highlights potential establishes foundation developing advanced diagnostic tools.

Language: Английский

---

A hybrid model for Arabic character recognition using CNN and Kolmogorov Arnold Networks (KANs)

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

Published: May 28, 2025

Language: Английский

---

Accurate breast cancer diagnosis strategy (BCDS) based on deep learning techniques

Neural Computing and Applications, Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 24, 2024

Language: Английский

---

Synergizing CNNs and Transformers for Accurate Face Age Estimation

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

Published: Dec. 2, 2024

Estimating the actual and perceived age of human faces has garnered significant interest for its wide-ranging practical applications. Various intelligent scenarios stand to gain from these computational systems capable accurately predicting individuals’ ages. Automated estimation are particularly valuable in fields such as medical diagnostics, facial product development, casting films, assessing impact cosmetic procedures, anti-aging treatments. In realm where deep networks have demonstrated their supremacy fron-trunners among machine learning tools, Our approach integrates Convolutional Neural Networks (CNN) with Transformers. This novel system enhances information extraction by utilizing transformer attention mechanisms, rather than solely depending on features extracted convolutional neural estimating age. Based experiments conducted, effectively captures sequential progression continuous nature aging process. Furthermore, proposed model surpasses cutting-edge delivering exceptional results, achieving lowest mean absolute errors 2.31 MORPH II, 5.35 CACD, 2.91 AFAD.

Language: Английский

---