Deep learning for the harmonization of structural MRI scans: a survey

BioMedical Engineering OnLine, Journal Year: 2024, Volume and Issue: 23(1)

Published: Aug. 31, 2024

Medical imaging datasets for research are frequently collected from multiple centers using different scanners, protocols, and settings. These variations affect data consistency compatibility across sources. Image harmonization is a critical step to mitigate the effects of factors like inherent differences between various vendors, hardware upgrades, protocol changes, scanner calibration drift, as well ensure consistent medical image processing techniques. Given importance widespread relevance this issue, vast array methodologies have emerged, with deep learning-based approaches driving substantial advancements in recent times. The goal review paper examine latest learning techniques employed by analyzing cutting-edge architectural field harmonization, evaluating both their strengths limitations. This begins providing comprehensive fundamental overview strategies, covering three aspects: established datasets, commonly used evaluation metrics, characteristics scanners. Subsequently, analyzes structural MRI (Magnetic Resonance Imaging) based on network architecture, algorithm, supervision strategy, output. underlying architectures include U-Net, Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), flow-based generative models, transformer-based approaches, custom-designed architectures. investigates effectiveness Disentangled Representation Learning (DRL) pivotal algorithm harmonization. Lastly, highlights primary limitations techniques, specifically lack quantitative comparisons methods. overall aim serve guide researchers practitioners select appropriate specific conditions requirements. It also aims foster discussions around ongoing challenges shed light promising future directions potential significant advancements.

Language: Английский

---

Automated liver and spleen segmentation for MR elastography maps using U-Nets

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

Published: March 28, 2025

To compare pretrained and trained U-Nets for liver spleen segmentation in multifrequency magnetic resonance elastography (MRE) magnitude images automated quantification of shear wave speed (SWS). Seventy-two healthy participants (34 ± 11 years; BMI, 23 2 kg/m2; 51 men) underwent MRE at 1.5T or 3T. Volumes interest (VOIs) were generated from with mixed T2-T2* image contrast then transferred to SWS maps. Pretrained 2D 3D compared ground truth values obtained by manual using correlation analysis, intraclass coefficients (ICCs), Dice scores. For both VOI values, pairwise comparison revealed no statistically significant difference between (all p ≥ 0.95). There was a strong positive R = 0.99 0.81-0.84 spleen. ICC 0.90-0.92 spleen, indicating excellent agreement good all investigated. scores showed performance networks the U-Net achieving slightly higher (0.95) (0.90), though differences three tested minimal. The we found when applying suggests that fully parameters within anatomical regions is feasible leveraging previously unexploited information conveyed images.

Language: Английский

---

Interpreting Survival Predictor Model for Glioblastoma Using Explainable Artificial Intelligence

Lecture notes in computer science, Journal Year: 2025, Volume and Issue: unknown, P. 74 - 91

Published: Jan. 1, 2025

Language: Английский

---

Edge-Adaptive Dynamic Scalable Convolution for Efficient Remote Mobile Pathology Analysis

ACM Transactions on Autonomous and Adaptive Systems, Journal Year: 2025, Volume and Issue: unknown

Published: April 28, 2025

With the emergence of edge computing, there’s a growing need for advanced technologies capable real-time, efficient processing complex data on devices, particularly in mobile health systems handling pathological images. On computing lightweighting models and reduction computational requirements not only save resources but also increase inference speed. Although many lightweight methods have been proposed recent years, they still face common challenges. This paper introduces novel convolution operation, Dynamic Scalable Convolution (DSC), which optimizes accelerates devices. DSC is shown to outperform traditional terms parameter efficiency, speed, overall performance, through comparative analyses computer vision tasks like image classification semantic segmentation. Experimental results demonstrate significant potential enhancing deep neural networks, applications smart devices remote healthcare, where it addresses challenge limited by reducing demands improving By integrating technology applications, offers promising approach support rapidly developing field, especially healthcare delivery multimedia communication.

Language: Английский

---

Advancements and gaps in natural language processing and machine learning applications in healthcare: a comprehensive review of electronic medical records and medical imaging

Frontiers in Physics, Journal Year: 2024, Volume and Issue: 12

Published: Dec. 2, 2024

This article presents a thorough examination of the progress and limitations in application Natural Language Processing (NLP) Machine Learning (ML), particularly Deep (DL), healthcare industry. paper examines utilisation (ML) field, specifically relation to Electronic Medical Records (EMRs). The review also incorporation medical imaging as supplementary emphasising transformative impact these technologies on analysis data patient care. attempts analyse both fields order offer insights into current state research suggest potential chances for future advancements. focus is use (EMRs) imaging. methodically detects, chooses, assesses literature published between 2015 2023, utilizing keywords pertaining natural language processing databases such SCOPUS. After applying precise inclusion criteria, 100 papers were thoroughly examined. emphasizes notable NLP ML methodologies improve decision-making, extract information from unorganized data, evaluate pictures. key findings highlight successful combination image enhance accuracy diagnoses study demonstrates effectiveness deep learning-based pipelines extracting valuable electronic records Additionally, suggests that has optimize allocation resources. identified gaps encompass necessity scalable practical implementations, improved interdisciplinary collaboration, consideration ethical factors, longitudinal customization approaches specific situations. Subsequent investigations should deficiencies fully exploit capabilities machine learning sector, consequently enhancing outcomes delivery services.

Language: Английский

---

Development and validation of CNN-MLP models for predicting anti-VEGF therapy outcomes in diabetic macular edema

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

Published: Dec. 4, 2024

Diabetic macular edema (DME) is a common complication of diabetes that can lead to vision loss, and anti-vascular endothelial growth factor (anti-VEGF) therapy the standard care for DME, but treatment outcomes vary widely among patients. This study collected optical coherence tomography (OCT) images clinical data from DME patients who received anti-VEGF develop validate deep learning (DL) models predicting in based on convolutional neural network (CNN) multilayer perceptron (MLP) combined architecture by using multimodal data. An Xception-MLP was utilized predict best-corrected visual acuity (BCVA), central subfield thickness (CST), cube volume (CV), average (CAT). Mean absolute error (MAE), mean squared (MSE) logarithmic (MSLE) were employed evaluate model performance. In this study, both training set validation exhibited consistent decreasing trend MAE, MSE, MSLE. No statistical difference found between actual predicted values all indicators. demonstrated improved CNN-MLP regression accurately BCVA, CST, CV, CAT after patients, which valuable ophthalmic decisions reduces economic burden

Language: Английский

---

YOLOv8-Seg: A Deep Learning Approach for Accurate Classification of Osteoporotic Vertebral Fractures

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

Published: July 17, 2024

The abstract of the article presents a study focused on application deep learning for classification osteoporotic vertebral fractures (OVF), growing health concern among elderly. research aimed to explore potential assist in diagnosing OVF, evaluate clinical viability this method, and enhance recovery rates. A dataset comprising 643 CT images OVF from patients admitted between March 2013 May 2023 was collected classified according European Vertebral Osteoporosis Study Group (EVOSG) spine system. Of these, 613 were utilized training validating model, while 30 served as test set assess model's performance against clinician diagnoses. system achieved an impressive 85.9% accuracy rate classifying EVOSG criteria. concludes that offers high degree identifying images, which could streamline improve current manual diagnostic process is often complex challenging. also introduces YOLOv8-Seg novel method designed capabilities OVF. use context positioned significant advancement with support medical professionals making early precise diagnoses, thereby improving patient outcomes. Key terms highlighted include learning, fracture, YOLOv8, indicating integration advanced technology diagnosis.

Language: Английский

---

Optimizing Multi Neural Network Weights for COVID-19 Detection Using Enhanced Artificial Ecosystem Algorithm

Traitement du signal, Journal Year: 2023, Volume and Issue: 40(4), P. 1491 - 1500

Published: Aug. 31, 2023

The role of machine learning in medical research, particularly addressing the COVID-19 pandemic, has proven to be significant.The current study delineates design and refinement an artificial intelligence (AI) framework tailored differentiate from Pneumonia utilizing X-ray scans synergy with textual clinical data.The focal point this research is amalgamation diverse neural networks exploration impact metaheuristic algorithms on optimizing these networks' weights.The proposed uniquely incorporates a lung segmentation process using pre-trained ResNet34 model, generating mask for each mitigate influence potential extraneous features.The dataset comprised 579 segmented images (Anteroposterior Posteroanterior views) patients, supplemented patient's data, including age gender.An enhancement accuracy 94.32% 97.85% was observed implementation weight optimization framework.The efficacy model detecting further ascertained through comprehensive comparison various architectures cited existing literature.

Language: Английский

---