Extended reality for biomedicine

Nature Reviews Methods Primers, Journal Year: 2023, Volume and Issue: 3(1)

Published: March 2, 2023

Language: Английский

---

Hierarchical Organ-Aware Total-Body Standard-Dose PET Reconstruction From Low-Dose PET and CT Images

IEEE Transactions on Neural Networks and Learning Systems, Journal Year: 2023, Volume and Issue: 35(10), P. 13258 - 13270

Published: May 10, 2023

Positron emission tomography (PET) is an important functional imaging technology in early disease diagnosis. Generally, the gamma ray emitted by standard-dose tracer inevitably increases exposure risk to patients. To reduce dosage, a lower dose often used and injected into However, this leads low-quality PET images. In article, we propose learning-based method reconstruct total-body (SPET) images from low-dose (LPET) corresponding computed (CT) Different previous works focusing only on certain part of human body, our framework can hierarchically SPET images, considering varying shapes intensity distributions different body parts. Specifically, first use one global network coarsely Then, four local networks are designed finely head-neck, thorax, abdomen-pelvic, leg parts body. Moreover, enhance each learning for respective part, design organ-aware with residual dynamic convolution (RO-DC) module dynamically adapting organ masks as additional inputs. Extensive experiments 65 samples collected uEXPLORER PET/CT system demonstrate that hierarchical consistently improve performance all parts, especially PSNR 30.6 dB, outperforming state-of-the-art methods image reconstruction.

Language: Английский

---

Federated Multi-Organ Segmentation With Inconsistent Labels

IEEE Transactions on Medical Imaging, Journal Year: 2023, Volume and Issue: 42(10), P. 2948 - 2960

Published: April 25, 2023

Federated learning is an emerging paradigm allowing large-scale decentralized without sharing data across different owners, which helps address the concern of privacy in medical image analysis. However, requirement for label consistency clients by existing methods largely narrows its application scope. In practice, each clinical site may only annotate certain organs interest with partial or no overlap other sites. Incorporating such partially labeled into a unified federation unexplored problem significance and urgency. This work tackles challenge using novel federated multi-encoding U-Net (Fed-MENU) method multi-organ segmentation. our method, (MENU-Net) proposed to extract organ-specific features through encoding sub-networks. Each sub-network can be seen as expert specific organ trained that client. Moreover, encourage extracted sub-networks informative distinctive, we regularize training MENU-Net designing auxiliary generic decoder (AGD). Extensive experiments on six public abdominal CT datasets show Fed-MENU effectively obtain model superior performance models either localized centralized methods. Source code publicly available at https://github.com/DIAL-RPI/Fed-MENU.

Language: Английский

---

Continual Segment: Towards a Single, Unified and Non-forgetting Continual Segmentation Model of 143 Whole-body Organs in CT Scans

2021 IEEE/CVF International Conference on Computer Vision (ICCV), Journal Year: 2023, Volume and Issue: unknown

Published: Oct. 1, 2023

Deep learning empowers the mainstream medical image segmentation methods. Nevertheless, current deep approaches are not capable of efficiently and effectively adapting updating trained models when new classes incrementally added. In real clinical environment, it can be preferred that could dynamically extended to segment organs/tumors without (re-)access previous training datasets due obstacles patient privacy data storage. This process viewed as a continual semantic (CSS) problem, being understudied for multi-organ segmentation. this work, we propose architectural CSS framework learn single model segmenting total 143 whole-body organs. Using encoder/decoder network structure, demonstrate continually then frozen encoder coupled with incrementally-added decoders extract sufficiently representative features subsequently validly segmented, while avoiding catastrophic forgetting in CSS. To maintain complexity, each decoder is progressively pruned using neural architecture search teacher-student based knowledge distillation. Finally, body-part anomaly-aware output merging module combine organ predictions originating from different incorporate both healthy pathological organs appearing datasets. Trained validated on 3D CT scans 2500+ patients four datasets, our very high accuracy, closely reaching upper bound performance level by separate (i.e., one per dataset/task).

Language: Английский

---

Multi-organ segmentation: a progressive exploration of learning paradigms under scarce annotation

Physics in Medicine and Biology, Journal Year: 2024, Volume and Issue: 69(11), P. 11TR01 - 11TR01

Published: March 13, 2024

Abstract Precise delineation of multiple organs or abnormal regions in the human body from medical images plays an essential role computer-aided diagnosis, surgical simulation, image-guided interventions, and especially radiotherapy treatment planning. Thus, it is great significance to explore automatic segmentation approaches, among which deep learning-based approaches have evolved rapidly witnessed remarkable progress multi-organ segmentation. However, obtaining appropriately sized fine-grained annotated dataset extremely hard expensive. Such scarce annotation limits development high-performance models but promotes many annotation-efficient learning paradigms. Among these, studies on transfer leveraging external datasets, semi-supervised including unannotated datasets partially-supervised integrating partially-labeled led dominant way break such dilemmas We first review fully supervised method, then present a comprehensive systematic elaboration 3 abovementioned paradigms context both technical methodological perspectives, finally summarize their challenges future trends.

Language: Английский

---

Deep-Learning for Rapid Estimation of the Out-of-Field Dose in External Beam Photon Radiation Therapy – A Proof of Concept

International Journal of Radiation Oncology*Biology*Physics, Journal Year: 2024, Volume and Issue: 120(1), P. 253 - 264

Published: March 28, 2024

The dose deposited outside of the treatment field during external photon beam radiation therapy treatment, also known as out-of-field dose, is subject extensive study it may be associated with a higher risk developing second cancer and could have deleterious effects on immune system that compromise efficiency combined radio-immunotherapy treatments. Out-of-field estimation tools developed today in research, including Monte Carlo simulations analytical methods, are not suited to requirements clinical implementation because their lack versatility cumbersome application. We propose proof concept based deep learning for map addresses these limitations.

Language: Английский

---

Exploration of anatomical distribution of brain metastasis from breast cancer at first diagnosis assisted by artificial intelligence

Heliyon, Journal Year: 2024, Volume and Issue: 10(9), P. e29350 - e29350

Published: April 18, 2024

ObjectivesThis study aimed to explore the spatial distribution of brain metastases (BMs) from breast cancer (BC) and identify high-risk sub-structures in BMs that are involved at first diagnosis.MethodsMagnetic resonance imaging (MRI) scans were retrospectively reviewed our centre. The was divided into eight regions according its anatomy function, volume each region calculated. identification calculation metastatic lesions accomplished using an automatically segmented 3D BUC-Net model. observed expected rates compared 2-tailed proportional hypothesis testing.ResultsA total 250 patients with BC who presented 1694 identified. overall incidences substructures as follows: cerebellum, 42.1%; frontal lobe, 20.1%; occipital 9.7%; temporal 8.0%; parietal 13.1%; thalamus, 4.7%; brainstem, 0.9%; hippocampus, 1.3%. Compared rate based on different regions, thalamus identified higher risk for (P value ≤ 5.6*10-3). Sub-group analysis type indicated triple-negative had a high involvement hippocampus brainstem.ConclusionsAmong BC, lobe higher-risk than BMs. brainstem areas triple negative cancer. However, further validation this conclusion requires larger sample size.

Language: Английский

---

Radiomics analysis of pancreas based on dual-energy computed tomography for the detection of type 2 diabetes mellitus

Frontiers in Medicine, Journal Year: 2024, Volume and Issue: 11

Published: April 19, 2024

Objective To utilize radiomics analysis on dual-energy CT images of the pancreas to establish a quantitative imaging biomarker for type 2 diabetes mellitus. Materials and methods In this retrospective study, 78 participants (45 with mellitus, 33 without) underwent dual energy exam. Pancreas regions were segmented automatically using deep learning algorithm. From these regions, features extracted. Additionally, 24 clinical collected each patient. Both then selected least absolute shrinkage selection operator (LASSO) technique build classifies random forest (RF), support vector machines (SVM) Logistic. Three models built: one features, combined model. Results Seven radiomic from while eight chosen pool LASSO method. These used model, its performance was evaluated five-fold cross-validation. The best classifier is Logistic reported area under curve (AUC) values test dataset 0.887 (0.73–1), 0.881 (0.715–1), 0.922 (0.804–1) respective models. Conclusion Radiomics offers potential as in detection

Language: Английский

---

Emerging pharmacotherapy trends in preventing and managing oral mucositis induced by chemoradiotherapy and targeted agents

Expert Opinion on Pharmacotherapy, Journal Year: 2024, Volume and Issue: 25(6), P. 727 - 742

Published: April 12, 2024

The introduction of targeted therapy and immunotherapy has tremendously changed the clinical outcomes prognosis cancer patients. Despite innovative pharmacological therapies improved radiotherapy (RT) techniques, patients continue to suffer from side effects, which oral mucositis (OM) is still most impactful, especially for quality life. We provide an overview current advances in pharmacotherapy RT, relation their potential cause OM, less explored more recent literature reports related best management OM. have analyzed natural/antioxidant agents, probiotics, mucosal protectants healing coadjuvants, pharmacotherapies, immunomodulatory anticancer photobiomodulation impact technology. discovery precise pathophysiologic mechanisms CT RT-induced OM outlined that a multifactorial origin, including direct oxidative damage, upregulation immunologic factors, effects on flora. A persistent upregulated immune response, associated with factors patients' characteristics, may contribute severe long-lasting goal strategies conjugate individual patient, disease, therapy-related guide prevention or treatment. further high-quality research warranted, issue paramount future strategies.

Language: Английский

---

Automatic detection of cognitive impairment in patients with white matter hyperintensity and causal analysis of related factors using artificial intelligence of MRI

Computers in Biology and Medicine, Journal Year: 2024, Volume and Issue: 178, P. 108684 - 108684

Published: June 4, 2024

Language: Английский

---