Guided Diffusion: Balancing Fidelity and Diversity in Dataset Augmentation with EfficientNet and Gaussian Noise

ITM Web of Conferences, Journal Year: 2025, Volume and Issue: 73, P. 02023 - 02023

Published: Jan. 1, 2025

The denoising diffusion probabilistic model (DDPM) has recently attracted massive attention due to its better capability of synthesizing high-quality and diverse synthetic data than generative adversarial network (GAN), paving the way for application in augmentation scenarios. However, balancing fidelity diversity remains a challenge. To address problem, novel architecture is proposed, incorporating EfficientNet extract features from original dataset fuse them with those noise samples, guiding process ensuring between samples data. Additionally, random Gaussian introduced UNet bottleneck output at each timestep enhance diversity. A pre-trained CNN classification follows ensure label consistency reference images. approach evaluated through experiments on lung cancer prediction using chest CT-scan dataset, achieving 13.6% improvement accuracy over baseline methods, 9.8% traditional cropping rotation approach, 4.1% GAN-based approach. These results validate effectiveness proposed method augmentation.

Language: Английский

---

Smart nanomedicines powered by artificial intelligence: a breakthrough in lung cancer diagnosis and treatment

Medical Oncology, Journal Year: 2025, Volume and Issue: 42(5)

Published: March 25, 2025

Language: Английский

---

Super perfect polarization-insensitive graphene disk terahertz absorber for breast cancer detection using deep learning

Optics & Laser Technology, Journal Year: 2024, Volume and Issue: 183, P. 112246 - 112246

Published: Dec. 8, 2024

Language: Английский

---

Lung Cancer Classification Using Deep Learning Hybrid Model

Advances in medical technologies and clinical practice book series, Journal Year: 2024, Volume and Issue: unknown, P. 207 - 223

Published: March 11, 2024

Abnormal growths in the lungs caused by disease. The classification of CT scans is accomplished applying machine learning strategies. Classification methods based on deep learning, such as support vector machines, can categorize a wide variety image datasets and produce segmentation results highest caliber. In this work, we suggested method for feature extraction from images altering SVM CNN then hybrid model resulting those modifications (NNSVLC). For investigation, Kaggle dataset will be utilized. proposed was found to accurate 91.7% time, determined experiments.

Language: Английский

---

A Review on Deep Learning for UAV Absolute Visual Localization

Drones, Journal Year: 2024, Volume and Issue: 8(11), P. 622 - 622

Published: Oct. 29, 2024

In the past few years, use of Unmanned Aerial Vehicles (UAVs) has expanded and now reached mainstream levels for applications such as infrastructure inspection, agriculture, transport, security, entertainment, real estate, environmental conservation, search rescue, even insurance. This surge in adoption can be attributed to UAV ecosystem’s maturation, which not only made these devices more accessible cost effective but also significantly enhanced their operational capabilities terms flight duration embedded computing power. conjunction with developments, research on Absolute Visual Localization (AVL) seen a resurgence driven by introduction deep learning field. These new approaches have improved localization solutions comparison previous generation based traditional computer vision feature extractors. paper conducts an extensive review literature learning-based methods AVL, covering significant advancements since 2019. It retraces key developments that led rise provides in-depth analysis related sources Inertial Measurement Units (IMUs) Global Navigation Satellite Systems (GNSSs), highlighting limitations advantages integration AVL. The concludes current challenges proposes future directions guide further work

Language: Английский

---

The application of Raman spectroscopy for the diagnosis and monitoring of lung tumors

Frontiers in Bioengineering and Biotechnology, Journal Year: 2024, Volume and Issue: 12

Published: April 18, 2024

Raman spectroscopy is an optical technique that uses inelastic light scattering in response to vibrating molecules produce chemical fingerprints of tissues, cells, and biofluids. strategies high levels specificity without requiring extensive sample preparation, allowing for the use advanced tools such as microscopes, fiber optics, lasers operate visible near-infrared spectral range, making them increasingly suitable a wide range medical diagnostic applications. Metal nanoparticles nonlinear effects can improve signals, optimized optic probes make real-time,

Language: Английский

---

Predicting COPD Readmission: An Intelligent Clinical Decision Support System

Diagnostics, Journal Year: 2025, Volume and Issue: 15(3), P. 318 - 318

Published: Jan. 29, 2025

Background: COPD is a chronic disease characterized by frequent exacerbations that require hospitalization, significantly increasing the care burden. In recent years, use of artificial intelligence-based tools to improve management patients with has progressed, but prediction readmission been less explored. fact, in state art, no models specifically designed make medium-term predictions (2–3 months after admission) have found. This work presents new intelligent clinical decision support system predict risk hospital 90 days an episode acute exacerbation. Methods: The structured two levels: first one consists three machine learning algorithms —Random Forest, Naïve Bayes, and Multilayer Perceptron—that operate concurrently readmission; second level, expert based on fuzzy inference engine combines generated risks, determining final prediction. employed database includes more than five hundred demographic, clinical, social variables. Prior building model, initial dataset was divided into training test subsets. order reduce high dimensionality problem, filter-based feature selection techniques were employed, followed recursive supported Random Forest algorithm, guaranteeing usability its potential integration environment. After knowledge base determined data subset using Wang–Mendel automatic rule generation algorithm. Results: Preliminary results obtained set are promising, AUC approximately 0.8. At selected cutoff point, sensitivity 0.67 specificity 0.75 achieved. Conclusions: highlights system’s future for early identification at readmission. For implementation practice, extensive validation process will be required, along expansion database, which likely contribute improving robustness generalization capacity.

Language: Английский

---

X-ray imaging-based lung cancer detection using SVM classifier

AIP conference proceedings, Journal Year: 2025, Volume and Issue: 3280, P. 030001 - 030001

Published: Jan. 1, 2025

Language: Английский

---

AI-Powered Lung Cancer Detection: Assessing VGG16 and CNN Architectures for CT Scan Image Classification

Informatics, Journal Year: 2025, Volume and Issue: 12(1), P. 18 - 18

Published: Feb. 11, 2025

Lung cancer is a leading cause of mortality worldwide, and early detection crucial in improving treatment outcomes reducing death rates. However, diagnosing medical images, such as Computed Tomography scans (CT scans), complex requires high level expertise. This study focuses on developing evaluating the performance Convolutional Neural Network (CNN) models, specifically Visual Geometry Group 16 (VGG16) architecture, to classify lung CT scan images into three categories: Normal, Benign, Malignant. The dataset used consists 1097 from 110 patients, categorized according these severity levels. research methodology began with data collection preparation, followed by training testing VGG16 model comparing its other CNN architectures, including Residual 50 layers (ResNet50), Inception Version 3 (InceptionV3), Mobile 2 (MobileNetV2). experimental results indicate that achieved highest classification performance, Test Accuracy 98.18%, surpassing models. accuracy highlights VGG16’s strong potential supportive diagnostic tool imaging. limitation this size, which may reduce when applied new data. Future studies should consider increasing using Data Augmentation techniques, fine-tuning parameters, employing advanced models 3D or Vision Transformers. Additionally, incorporating Gradient-weighted Class Activation Mapping (Grad-CAM) interpret decisions would enhance transparency reliability. confirms CNNs, particularly VGG16, for classifying provides foundation further development applications.

Language: Английский

---

SALM: A Unified Model for 2D and 3D Region of Interest Segmentation in Lung CT Scans Using Vision Transformers

Applied Biosciences, Journal Year: 2025, Volume and Issue: 4(1), P. 11 - 11

Published: Feb. 17, 2025

Accurate segmentation of Regions Interest (ROI) in lung Computed Tomography (CT) is crucial for early cancer diagnosis and treatment planning. However, the variability size, shape, location lesions, along with complexity 3D spatial relationships, poses significant challenges. In this work, we propose SALM (Segment Anything Lung Model), a deep learning model 2D ROI segmentation. leverages Vision Transformers, proposing an adaptation positional encoding functions to effectively capture relationships both slices volumes using single, unified model. Evaluation on LUNA16 dataset demonstrated strong performance modalities. segmentation, achieved Dice score 93% 124,662 slices. For 174 images from same dataset, attained 81.88%. We also tested external database (PleThora) subset 255 pulmonary CT diseased patients, where it 78.82%. These results highlight SALM’s ability accurately segment 3D, demonstrating its potential improve accuracy efficiency computer-aided cancer.

Language: Английский

---