Advancements in cardiac structures segmentation: a comprehensive systematic review of deep learning in CT imaging

Frontiers in Cardiovascular Medicine, Год журнала: 2024, Номер 11

Опубликована: Янв. 22, 2024

Background Segmentation of cardiac structures is an important step in evaluation the heart on imaging. There has been growing interest how artificial intelligence (AI) methods—particularly deep learning (DL)—can be used to automate this process. Existing AI approaches segmentation have mostly focused MRI. This systematic review aimed appraise performance and quality supervised DL tools for CT. Methods Embase Medline databases were searched identify related studies from January 1, 2013 December 4, 2023. Original research published peer-reviewed journals after eligible inclusion if they presented DL-based non-coronary great vessels The data extracted included information about structure(s) being segmented, study location, architectures reported metrics such as Dice similarity coefficient (DSC). was assessed using Checklist Artificial Intelligence Medical Imaging (CLAIM). Results 18 2020 included. DSC scores median achieved most commonly segmented left atrium (0.88, IQR 0.83–0.91), ventricle (0.91, 0.89–0.94), myocardium (0.83, 0.82–0.92), right 0.83–0.90), 0.85–0.92), pulmonary artery (0.92, 0.87–0.93). Compliance with CLAIM variable. In particular, only 58% showed compliance dataset description criteria did not test or validate their models external (81%). Conclusion Supervised applied various Most similar measured by values. limited size nature training datasets, inconsistent descriptions ground truth annotations lack testing clinical settings. Systematic Review Registration [ www.crd.york.ac.uk/prospero/ ], PROSPERO [CRD42023431113].

Язык: Английский

---

Delineation of clinical target volume and organs at risk in cervical cancer radiotherapy by deep learning networks

Medical Physics, Год журнала: 2023, Номер 50(10), С. 6354 - 6365

Опубликована: Май 29, 2023

Delineation of the clinical target volume (CTV) and organs-at-risk (OARs) is important in cervical cancer radiotherapy. But it generally labor-intensive, time-consuming, subjective. This paper proposes a parallel-path attention fusion network (PPAF-net) to overcome these disadvantages delineation task.The PPAF-net utilizes both texture structure information CTV OARs by employing U-Net capture high-level information, an up-sampling down-sampling (USDS) low-level accentuate boundaries OARs. Multi-level features extracted from networks are then fused together through module generate result.The dataset contains 276 computed tomography (CT) scans patients with staging IB-IIA. The images provided West China Hospital Sichuan University. Simulation results demonstrate that performs favorably on (e.g., rectum, bladder etc.) achieves state-of-the-art accuracy, respectively, for In terms Dice Similarity Coefficient (DSC) Hausdorff Distance (HD), 88.61% 2.25 cm CTV, 92.27% 0.73 96.74% 0.68 bladder, 96.38% 0.65 left kidney, 96.79% 0.63 right 93.42% 0.52 femoral head, 93.69% 0.51 87.53% 1.07 small intestine, 91.50% 0.84 spinal cord.The proposed automatic well segmentation tasks, which has great potential reducing burden radiation oncologists increasing accuracy delineation. future, University will further evaluate delineation, making this method helpful practice.

Язык: Английский

---

Breomics-U2net: Intra-Stage Multi-Scale Dual-Stream Atrous Convolutional Neural Network with Efficient Channel Attention-Based Fusion for Breast Cancer Segmentation Using Automated Breast Ultrasound (Abus)

Опубликована: Янв. 1, 2025

Язык: Английский

---

Detection and segmentation of pulmonary embolism in 3D CT pulmonary angiography using a threshold adjustment segmentation network

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Март 1, 2025

Язык: Английский

---

Path‐enhanced chunking approach with residual attention for medical image segmentation

Medical Physics, Год журнала: 2025, Номер unknown

Опубликована: Март 8, 2025

Medical image segmentation is an essential component of computer-aided diagnosis. While U-Net has been widely used in this field, its performance can be limited by incomplete feature information transfer and the imbalance between foreground background pixel classes medical images. To improve utilization address challenges, such as missing target regions insufficient edge detail preservation, study proposes a method that integrates path enhancement, residual attention, zone-based chunking training. The proposed introduces enhancement structure consisting bottom-up aggregation branch (PAB) multilevel fusion complementary (FEB). PAB aims to transmission semantic positional information, while FEB provides richer representation for mask prediction. Additionally, block with directional frontier support combinatorial attention designed focus on important content units boundary features. further refine segmentation, strategy employed enhance extraction fine-grained details through localized processing. was evaluated extensive ablation experiments, demonstrating consistent across multiple trials. When applied lung nodule computed tomography (CT) images, showed reduction mis-segmented regions. experimental results suggest approach accuracy stability compared baseline methods. Overall, shows promise tasks, particularly applications requiring precise delineation complex structures.

Язык: Английский

---

Few-shot small vessel segmentation using a detail-preserving network enhanced by discriminator

Medical & Biological Engineering & Computing, Год журнала: 2025, Номер unknown

Опубликована: Май 13, 2025

Язык: Английский

---

Applications of artificial intelligence in computed tomography imaging for phenotyping pulmonary hypertension

Current Opinion in Pulmonary Medicine, Год журнала: 2024, Номер 30(5), С. 464 - 472

Опубликована: Июль 9, 2024

Purpose of review Pulmonary hypertension is a heterogeneous condition with significant morbidity and mortality. Computer tomography (CT) plays central role in determining the phenotype pulmonary hypertension, informing treatment strategies. Many artificial intelligence tools have been developed this modality for assessment hypertension. This article reviews latest CT applications related diseases. Recent findings Multistructure segmentation both nonpulmonary cohorts using state-of-the-art UNet architecture. These segmentations correspond well those trained radiologists, giving clinically valuable metrics significantly less time. Artificial lung parenchymal accurately identifies quantifies disease patterns by integrating multiple radiomic techniques such as texture analysis classification. gives information on burden prognosis. There are many accurate to detect acute embolism. Detection chronic embolism proves more challenging further research required. Summary numerous being identify quantify relevant parameters cohorts. potentially provide efficient clinical information, impacting decision-making.

Язык: Английский

---

SPE-YOLO: A deep learning model focusing on small pulmonary embolism detection

Computers in Biology and Medicine, Год журнала: 2024, Номер 184, С. 109402 - 109402

Опубликована: Ноя. 13, 2024

Язык: Английский

---

A dual-task region-boundary aware neural network for accurate pulmonary nodule segmentation

Journal of Visual Communication and Image Representation, Год журнала: 2023, Номер 96, С. 103909 - 103909

Опубликована: Авг. 7, 2023

Язык: Английский

---

Heart and great vessels segmentation in congenital heart disease via CNN and conditioned energy function postprocessing

International Journal of Computer Assisted Radiology and Surgery, Год журнала: 2024, Номер 19(8), С. 1597 - 1605

Опубликована: Май 30, 2024

Язык: Английский

---