MRI Segmentation Research Articles

Comprehensive analysis of synthetic learning applied to neonatal brain MRI segmentation.

Brain segmentation from neonatal MRI images is a very challenging task due to large changes in the shape of cerebral structures and variations in signal intensities reflecting the gestational process. In this context, there is a clear need for segmentation techniques that are robust to variations in image contrast and to the spatial configuration of anatomical structures. In this work, we evaluate the potential of synthetic learning, a contrast-independent model trained using synthetic images generated from the ground truth labels of very few subjects. We base our experiments on the dataset released by the developmental Human Connectome Project, for which high-quality images are available for more than 700 babies aged between 26 and 45 weeks postconception. First, we confirm the impressive performance of a standard UNet trained on a few volumes, but also confirm that such models learn intensity-related features specific to the training domain. We then confirm the robustness of the synthetic learning approach to variations in image contrast. However, we observe a clear influence of the age of the baby on the predictions. We improve the performance of this model by enriching the synthetic training set with realistic motion artifacts and over-segmentation of the white matter. Based on extensive visual assessment, we argue that the better performance of the model trained on real T2w data may be due to systematic errors in the ground truth. We propose an original experiment allowing us to show that learning from real data will reproduce any systematic bias affecting the training set, while synthetic models can avoid this limitation. Overall, our experiments confirm that synthetic learning is an effective solution for segmenting neonatal brain MRI. Our adapted synthetic learning approach combines key features that will be instrumental for large multisite studies and clinical applications.

Automatic design of W-operators using membership functions: a case study in brain MRI segmentation

A W-operator is an image transformation that is locally defined inside a window W, invariant to translations. The automatic design of the W-operators consists of the design of functions, whose domain is a set of patterns or vectors obtained by translating a window through training images and the output of each vector is a class or label. The main difficulty to consider when designing W-operators is the generalization problem that occurs due to lack of training images. In this work, we propose the use of membership functions to solve the generalization problem in gray level images. Membership functions are defined from the training images to model regions that are often inaccurate due to ambiguous gray levels in the images. This proposal was applied to brain magnetic resonance image segmentation to test its performance in a field of interest in biomedical images. The experiments were carried out with different numbers of training and test images, windows sizes of 3×3\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$3\ imes 3$$\\end{document}, 5×5\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$5\ imes 5$$\\end{document}, 7×7\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$7\ imes 7$$\\end{document}, 11×11\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$11\ imes 11$$\\end{document}, and 15×15\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$15\ imes 15$$\\end{document}, and images with noise levels at 0, 1, 3, 5, 7, and 9%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document}. To calculate the performance of each designed W-operator, the classification error, sensitivity, and specificity were used. From the experimental results, it was concluded that the best performance is achieved with a window of size 3×3\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$3\ imes 3$$\\end{document}. In images with noise levels from 1 to 5%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document}, the classification error is less than 4%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document} and the sensitivity and specificity are greater than 94 and 98%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document}, respectively.

Quantitative Comparison of Liver Volume, Proton Density Fat Fraction, and Time Burden between Automatic Whole Liver Segmentation and Manual Sampling MRI Strategies for Diagnosing Metabolic Dysfunction-associated Steatotic Liver Disease in Obese Patients.

The performance of automatic liver segmentation and manual sampling MRI strategies needs be compared to determine interchangeability. To compare automatic liver segmentation and manual sampling strategies (manual whole liver segmentation and standardized manual region of interest) for performance in quantifying liver volume and MRI-proton density fat fraction (MRI-PDFF), identifying steatosis grade, and time burden. Fifty patients with obesity who underwent liver biopsy and MRI between December 2017 and November 2018 were included. Sampling strategies included automatic and manual whole liver segmentation and 4 and 9 large regions of interest. Intraclass correlation coefficient (ICC), Bland-Altman, linear regression, receiver operating characteristic curve, and Pearson correlation analyses were performed. Automatic whole liver segmentation liver volume and manual whole liver segmentation liver volume showed excellent agreement (ICC=0.97), high correlation (R2=0.96), and low bias (3.7%, 95% limits of agreement, -4.8%, 12.2%) in liver volume. There was the best agreement (ICC=0.99), highest correlation (R2=1.00), and minimum bias (0.84%, 95% limits of agreement, -0.20%, 1.89%) between automated whole liver segmentation MRI-PDFF and manual whole liver segmentation MRI-PDFF. There was no difference of each paired comparison of receiver operating characteristic curves for detecting steatosis (P=0.07-1.00). The minimum time burden for automatic whole liver segmentation was 0.32 s (0.32-0.33 s). Automatic measurement has similar effects to manual measurement in quantifying liver volume, MRI-PDFF, and detecting steatosis. Time burden of automatic whole liver segmentation is minimal among all sampling strategies. Manual measurement can be replaced by automatic measurement to improve quantitative efficiency.

Usformer: A small network for left atrium segmentation of 3D LGE MRI

Left atrial (LA) fibrosis plays a vital role as a mediator in the progression of atrial fibrillation. 3D late gadolinium-enhancement (LGE) MRI has been proven effective in identifying LA fibrosis. Image analysis of 3D LA LGE involves manual segmentation of the LA wall, which is both lengthy and challenging. Automated segmentation poses challenges owing to the diverse intensities in data from various vendors, the limited contrast between LA and surrounding tissues, and the intricate anatomical structures of the LA. Current approaches relying on 3D networks are computationally intensive since 3D LGE MRIs and the networks are large. Regarding this issue, most researchers came up with two-stage methods: initially identifying the LA center using a scaled-down version of the MRIs and subsequently cropping the full-resolution MRIs around the LA center for final segmentation. We propose a lightweight transformer-based 3D architecture, Usformer, designed to precisely segment LA volume in a single stage, eliminating error propagation associated with suboptimal two-stage training. The transposed attention facilitates capturing the global context in large 3D volumes without significant computation requirements. Usformer outperforms the state-of-the-art supervised learning methods in terms of accuracy and speed. First, with the smallest Hausdorff Distance (HD) and Average Symmetric Surface Distance (ASSD), it achieved a dice score of 93.1% and 92.0% in the 2018 Atrial Segmentation Challenge and our local institutional dataset, respectively. Second, the number of parameters and computation complexity are largely reduced by 2.8x and 3.8x, respectively. Moreover, Usformer does not require a large dataset. When only 16 labeled MRI scans are used for training, Usformer achieves a 92.1% dice score in the challenge dataset. The proposed Usformer delineates the boundaries of the LA wall relatively accurately, which may assist in the clinical translation of LA LGE for planning catheter ablation of atrial fibrillation.

Optimising total knee replacement imaging: a novel 3D printed PET/CT anthropomorphic phantom for metal artefact simulation

PurposeArthroplasty phantoms, including total knee replacement (TKR) phantoms, have been frequently used to test metal artefact reduction methods applied to positron emission tomography/computed tomography (PET/CT) images. These phantoms generally simulate either simple anatomical features or simple activity distribution around the metal inserts in the PET/CT scans. 3D printing has been used recently to fabricate fillable anthropomorphic phantoms that accurately simulate volume and geometry. This study aims to describe the process of image segmentation, phantom modelling, 3D printing and validation of a population-based fillable TKR phantom that simulates human TKR PET/CT metal artefacts.Methods10 participants (5 male and 5 female) were scanned using 3T MRI and the images were segmented to create average male and average female 3D knee models, inversely with void cortical and porous trabecular compartments for 3D printing and contrast media. Virtual total knee replacement (TKR) surgery was implemented on these models to prepare the insertion locations for knee prosthetic implants. Subsequently, TKR models were printed using a 3D photopolymer resin printer and then injected with normal saline to test the phantoms for any leaks. Subsequently, diluted iodinated contrast media was injected into the cortical compartment and saline with 18F-FDG was injected into the trabecular compartment and the phantom was scanned with PET/CT. The images were then evaluated and compared to the human knee radiographic features reported in the literature.ResultsPhantoms were shown to be fluid-tight with distinct compartments. They showed comparable volume and geometry to the segmented human MRI knees. The phantoms demonstrated similar values for x-ray attenuation and Hounsfield units (HU) to the literature for both cortical and trabecular compartments. The phantoms displayed a uniform distribution for the radioactive tracer, resembling that seen in human trabecular bone PET. TKR phantom PET/CT images with metal inserts replicated the clinical metal artefacts seen clinically in the periprosthetic area.ConclusionThis novel, 3D-printed, and customisable phantom effectively mimics the geometric, radiographic and radiotracer distribution features of real TKRs. Importantly, it simulates TKR image metal artefacts, making it suitable for repeatable and comprehensive evaluation of various metal artefact reduction methods in future research.

DECIDE: A decoupled semantic and boundary learning network for precise osteosarcoma segmentation by integrating multi-modality MRI

Automated Osteosarcoma Segmentation in Multi-modality MRI (AOSMM) holds clinical significance for effective tumor evaluation and treatment planning. However, the precision of AOSMM is challenged by the diverse characteristics of multi-modality MRI and the inherent heterogeneity and boundary ambiguity of osteosarcoma. While numerous methods have made significant strides in automated osteosarcoma segmentation, they primarily focused on the use of a single MRI modality and overlooked the potential benefits of integrating complementary information from other MRI modalities. Furthermore, they did not adequately model the long-range dependencies of complex tumor features, which may lead to insufficiently discriminative feature representations. To this end, we propose a decoupled semantic and boundary learning network (DECIDE) to achieve precise AOSMM with three functional modules. The Multi-modality Feature Fusion and Recalibration (MFR) module adaptively fuses and recalibrates multi-modality features by exploiting their channel-wise dependencies to compute low-rank attention weights for effectively aggregating useful information from different MRI modalities, which promotes complementary learning between multi-modality MRI and enables a more comprehensive tumor characterization. The Lesion Attention Enhancement (LAE) module employs spatial and channel attention mechanisms to capture global contextual dependencies over local features, significantly enhancing the discriminability and representational capacity of intricate tumor features. The Boundary Context Aggregation (BCA) module further enhances semantic representations by utilizing boundary information for effective context aggregation while also ensuring intra-class consistency in cases of boundary ambiguity. Substantial experiments demonstrate that DECIDE achieves exceptional performance in osteosarcoma segmentation, surpassing state-of-the-art methods in terms of accuracy and stability.

A Comprehensive Brain MRI Image Segmentation System Based on Contourlet Transform and Deep Neural Networks

Brain tumors are one of the deadliest types of cancer. Rapid and accurate identification of brain tumors, followed by appropriate surgical intervention or chemotherapy, increases the probability of survival. Accurate determination of brain tumors in MRI scans determines the exact location of surgical intervention or chemotherapy. However, this accurate segmentation of brain tumors, due to their diverse morphologies in MRI scans, poses challenges that require significant expertise and accuracy in image interpretation. Despite significant advances in this field, there are several barriers to proper data collection, particularly in the medical sciences, due to concerns about the confidentiality of patient information. However, research papers for learning systems and proposed networks often rely on standardized datasets because a specific approach is unavailable. This system combines unsupervised learning in the adversarial generative network component with supervised learning in segmentation networks. The system is fully automated and can be applied to tumor segmentation on various datasets, including those with sparse data. In order to improve the learning process, the brain MRI segmentation network is trained using a generative adversarial network to increase the number of images. The U-Net model was employed during the segmentation step to combine the remaining blocks efficiently. Contourlet transform produces the ground truth for each MRI image obtained from the adversarial generator network and the original images in the processing and mask preparation phase. On the part of the adversarial generator network, high-quality images are produced, the results of which are similar to the histogram of the original images. Finally, this system improves the image segmentation performance by combining the remaining blocks with the U-net network. Segmentation is evaluated using brain magnetic resonance images obtained from Istanbul Medipol Hospital. The results show that the proposed method and image segmentation network, which incorporates several criteria, such as the DICE criterion of 0.9434, can be effectively used in any dataset as a fully automatic system for segmenting different brain MRI images.

Automated body organ segmentation, volumetry and population-averaged atlas for 3D motion-corrected T2-weighted fetal body MRI

Structural fetal body MRI provides true 3D information required for volumetry of fetal organs. However, current clinical and research practice primarily relies on manual slice-wise segmentation of raw T2-weighted stacks, which is time consuming, subject to inter- and intra-observer bias and affected by motion-corruption. Furthermore, there are no existing standard guidelines defining a universal approach to parcellation of fetal organs. This work produces the first parcellation protocol of the fetal body organs for motion-corrected 3D fetal body MRI. It includes 10 organ ROIs relevant to fetal quantitative volumetry studies. We also introduce the first population-averaged T2w MRI atlas of the fetal body. The protocol was used as a basis for training of a neural network for automated organ segmentation. It showed robust performance for different gestational ages. This solution minimises the need for manual editing and significantly reduces time. The general feasibility of the proposed pipeline was also assessed by analysis of organ growth charts created from automated parcellations of 91 normal control 3T MRI datasets that showed expected increase in volumetry during 22–38 weeks gestational age range.

Fast and accurate 3-D spine MRI segmentation using FastCleverSeg

Accurate and efficient segmenting of vertebral bodies, muscles, and discs is crucial for analyzing various spinal diseases. However, traditional methods are either laborious and time-consuming (manual segmentation) or require extensive training data (fully automatic segmentation). FastCleverSeg, our proposed semi-automatic segmentation approach, addresses those limitations by significantly reducing user interaction while maintaining high accuracy. First, we reduce user interaction by requiring the manual annotation of only two or three slices. Next, we automatically Estimate the Annotation on Intermediary Slices (EANIS) using traditional computer vision/graphics concepts. Finally, our proposed method leverages improved voxel weight balancing to achieve fast and precise volumetric segmentation in the segmentation process. Experimental evaluations on our assembled diverse MRI databases comprising 179 patients (60 male, 119 female), demonstrate a remarkable 25 ms (30 ms standard deviation) processing time and a significant reduction in user interaction compared to existing approaches. Importantly, FastCleverSeg maintains or surpasses the segmentation quality of competing methods, achieving a Dice score of 94%. This invaluable tool empowers physicians to efficiently generate reliable ground truths, expediting the segmentation process and paving the way for future integration with deep learning approaches. In turn, this opens exciting possibilities for future fully automated spine segmentation.

Single brain metastases - prognostic factors and impact of residual tumor burden on overall survival.

Brain metastases (BM) are a common and challenging issue, with their incidence on the rise due to advancements in systemic therapies and increased patient survival. Most patients present with single BM, some of them without any further extracranial metastasis (i.e., solitary BM). The significance of postoperative intracranial tumor volume in the treatment of singular and solitary BM is still debated. This study aimed to determine the impact of resection and postoperative tumor burden on overall survival (OS) in patients with single BM. Patients with surgically treated single BM between 04/2007-01/2020 were retrospectively included. Residual tumor burden (RTB) was determined by manual segmentation of early postoperative brain MRI (72 h). Survival analyses were performed using Kaplan-Meier estimates for univariate analysis and Cox regression proportional hazards model for multivariate analysis, using preoperative Karnofsky performance status scale (KPSS), age, sex, RTB, incomplete resection and singular/solitary BM as covariates. 340 patients were included, median age 64 years (54-71). 119 patients (35%) had solitary BM, 221 (65%) singular BM. Complete resection (RTB=0) was achieved in 73%, median preoperative tumor burden was 11.2 cm3 (5-25), and RTB 0 cm3 (0-0.2). Median OS of patients with singular BM was 13 months (4-33) vs 20 months (5-92) for solitary BM; p=0.062. Multivariate analysis revealed singular BM as independent risk factor for poorer OS: HR 1.840 (1.202-2.817), p=0.005. Complete vs. incomplete resection showed no significant OS difference (13 vs. 13 months, p=0.737). When focusing on solitary BM, complete resection led to a longer OS than incomplete resection (21 vs. 8 months), without statistical significance(p=0.250). Achieving RTB=0 resulted in higher OS for patients with solitary BM compared to singular BM (21 vs. 12 months, p=0.027). Patients who received postoperative radiotherapy (RT) had significantly longer OS compared to those without it (14 vs. 4 months, p<0.001), with favorable OS in those receiving stereotactic radiosurgery (SRS) (15 months (3-42), p<0.001) or hypofractionated stereotactic radiotherapy (HSRT). When complete intracranial tumor resection RTB=0 is achieved, patients with solitary BM have a favorable outcome compared to singular BM. Singular BM was confirmed as independent risk factor. There is a strong presumption that complete resection leads to an improved oncological prognosis. Patients with solitary BM tend to benefit with a favorable outcome following complete resection. Hence, surgical resection should be considered as a treatment option for patients presenting with either no or minimal extracranial disease. Furthermore, the highly favorable impact of postoperative RT on OS was demonstrated and confirmed, especially with SRS or HSRT.

Possibilistic exponential spatial fuzzy clustering based cancer segmentation in multi-parametric prostate MRI

Possibilistic exponential spatial fuzzy clustering based cancer segmentation in multi-parametric prostate MRI

Early stage tumor segmentation in breast MRI using shape enhanced U-Net

Breast cancer is one of the most common cancers in women. Early diagnosis using MRI can significantly increase the cure rate. However, breast tumor in early stage is small whose edge is ambiguous, leading to false detection and missing detection. To this end, we propose a shape enhanced U-Net with Transformer encoder layer (called SETE-UNet) for automatically segmenting the tumor from breast cancer. First, Transformer encoder layer is introduced to provide global self-attention mechanisms, achieving obvious promotion for tumor segmentation. Second, a discriminative representation is calculated by the proposed shape enhanced branch, which introduces a novel branch to learn a shape boundary with additional supervision. This proposed module can help our algorithm focus on the relevant boundary and also achieve further performance improvement for tiny tumor. Furthermore, we construct an MRI dataset of breast tumors in early stage for breast tumor segmentation, containing 260 cases, including 243 women and 17 men. We compare the proposed method with the popular algorithms on our dataset of breast cancer in early stage, which shows that our method presents excellent segmentation results on breast MRI containing various organs with respect to MIoU, Precision, Accuracy, Dice, Sensitivity, Specificity, AUC, Hausdorff Distance.

Deep learning-based tumor segmentation and classification in breast MRI with 3TP method

Background and Objective:Timely diagnosis of early breast cancer plays a critical role in improving patient outcome and increasing treatment effectiveness. Dynamic contrast-enhancing magnetic resonance imaging (DCE-MRI) is a minimally invasive test widely used in the analysis of breast cancer. Manual analysis of DCE-MRI images by the specialist is extremely complex, exhaustive, and can lead to misunderstandings. Thus, the development of automated methods for analyzing DCE-MRI images of the breast is increasing. In this research, we propose an automatic methodology capable of detecting tumors and classifying their malignancy in a DCE-MRI breast image. Methodology:The proposed method consists of the use of two deep learning architectures, that is, SegNet and UNet, for breast tumor segmentation and the three-time-point (3TP) method for classifying the malignancy of segmented tumors. Results:The proposed methodology was tested on the public Quantitative Imaging Network (QIN) Breast DCE-MRI image set, and the best result in segmentation was a Dice of 0.9332 and IoU of 0.9799. For the classification of tumor malignancy, the methodology presented an accuracy of 100%. Conclusions:In our research, we demonstrate that the problem of mammary tumor segmentation in DCE-MRI images can be efficiently solved using deep learning architectures, and tumor malignancy classification can be done through the three-time method. The method can be integrated as a support system for the specialist in treating patients with breast cancer.

Segmentation of brain MRI using moth-flame optimization with modified cross entropy based fitness function

Segmentation of brain MRI using moth-flame optimization with modified cross entropy based fitness function

Auto-segmentation of Adult-Type Diffuse Gliomas: Comparison of Transfer Learning-Based Convolutional Neural Network Model vs. Radiologists

Segmentation of glioma is crucial for quantitative brain tumor assessment, to guide therapeutic research and clinical management, but very time-consuming. Fully automated tools for the segmentation of multi-sequence MRI are needed. We developed and pretrained a deep learning (DL) model using publicly available datasets A (n = 210) and B (n = 369) containing FLAIR, T2WI, and contrast-enhanced (CE)-T1WI. This was then fine-tuned with our institutional dataset (n = 197) containing ADC, T2WI, and CE-T1WI, manually annotated by radiologists, and split into training (n = 100) and testing (n = 97) sets. The Dice similarity coefficient (DSC) was used to compare model outputs and manual labels. A third independent radiologist assessed segmentation quality on a semi-quantitative 5-scale score. Differences in DSC between new and recurrent gliomas, and between uni or multifocal gliomas were analyzed using the Mann–Whitney test. Semi-quantitative analyses were compared using the chi-square test. We found that there was good agreement between segmentations from the fine-tuned DL model and ground truth manual segmentations (median DSC: 0.729, std-dev: 0.134). DSC was higher for newly diagnosed (0.807) than recurrent (0.698) (p < 0.001), and higher for unifocal (0.747) than multi-focal (0.613) cases (p = 0.001). Semi-quantitative scores of DL and manual segmentation were not significantly different (mean: 3.567 vs. 3.639; 93.8% vs. 97.9% scoring ≥ 3, p = 0.107). In conclusion, the proposed transfer learning DL performed similarly to human radiologists in glioma segmentation on both structural and ADC sequences. Further improvement in segmenting challenging postoperative and multifocal glioma cases is needed.

Quantifying model uncertainty for semantic segmentation of Fluorine-19 MRI using stochastic gradient MCMC

Fluorine-19 (19F) MRI is an emerging theranostic tool for studying diseases and treatments simultaneously, particularly in challenging neuroinflammatory conditions. However, the low signal-to-noise ratio (SNR) of 19F MRI necessitates computational methods to reliably detect 19F signal regions and segment these from the background. In this study, we demonstrate that Bayesian fully convolutional neural networks provide a means to increase sensitivity in 19F MRI and simultaneously provide estimates of data uncertainty. While our model effectively denoises the data, uncertain areas remain, particularly in boundary regions of the foreground. The uncertainty estimates are beneficial in preventing overconfident downstream analysis on noisy data and providing crucial information for rectifying prediction errors. Our results demonstrate that our model significantly outperforms other commonly used methods for 19F MRI signal detection in terms of sensitivity, while also providing valuable uncertainty estimates.

Assessment of deep learning segmentation for real-time free-breathing cardiac magnetic resonance imaging at rest and under exercise stress

In recent years, a variety of deep learning networks for cardiac MRI (CMR) segmentation have been developed and analyzed. However, nearly all of them are focused on cine CMR under breathold. In this work, accuracy of deep learning methods is assessed for volumetric analysis (via segmentation) of the left ventricle in real-time free-breathing CMR at rest and under exercise stress. Data from healthy volunteers (n = 15) for cine and real-time free-breathing CMR at rest and under exercise stress were analyzed retrospectively. Exercise stress was performed using an ergometer in the supine position. Segmentations of two deep learning methods, a commercially available technique (comDL) and an openly available network (nnU-Net), were compared to a reference model created via the manual correction of segmentations obtained with comDL. Segmentations of left ventricular endocardium (LV), left ventricular myocardium (MYO), and right ventricle (RV) are compared for both end-systolic and end-diastolic phases and analyzed with Dice’s coefficient. The volumetric analysis includes the cardiac function parameters LV end-diastolic volume (EDV), LV end-systolic volume (ESV), and LV ejection fraction (EF), evaluated with respect to both absolute and relative differences. For cine CMR, nnU-Net and comDL achieve Dice’s coefficients above 0.95 for LV and 0.9 for MYO, and RV. For real-time CMR, the accuracy of nnU-Net exceeds that of comDL overall. For real-time CMR at rest, nnU-Net achieves Dice’s coefficients of 0.94 for LV, 0.89 for MYO, and 0.90 for RV and the mean absolute differences between nnU-Net and the reference are 2.9 mL for EDV, 3.5 mL for ESV, and 2.6% for EF. For real-time CMR under exercise stress, nnU-Net achieves Dice’s coefficients of 0.92 for LV, 0.85 for MYO, and 0.83 for RV and the mean absolute differences between nnU-Net and reference are 11.4 mL for EDV, 2.9 mL for ESV, and 3.6% for EF. Deep learning methods designed or trained for cine CMR segmentation can perform well on real-time CMR. For real-time free-breathing CMR at rest, the performance of deep learning methods is comparable to inter-observer variability in cine CMR and is usable for fully automatic segmentation. For real-time CMR under exercise stress, the performance of nnU-Net could promise a higher degree of automation in the future.

Navigating the nuances: comparative analysis and hyperparameter optimisation of neural architectures on contrast-enhanced MRI for liver and liver tumour segmentation

In medical imaging, accurate segmentation is crucial to improving diagnosis, treatment, or both. However, navigating the multitude of available architectures for automatic segmentation can be overwhelming, making it challenging to determine the appropriate type of architecture and tune the most crucial parameters during dataset optimisation. To address this problem, we examined and refined seven distinct architectures for segmenting the liver, as well as liver tumours, with a restricted training collection of 60 3D contrast-enhanced magnetic resonance images (CE-MRI) from the ATLAS dataset. Included in these architectures are convolutional neural networks (CNNs), transformers, and hybrid CNN/transformer architectures. Bayesian search techniques were used for hyperparameter tuning to hasten convergence to the optimal parameter mixes while also minimising the number of trained models. It was unexpected that hybrid models, which typically exhibit superior performance on larger datasets, would exhibit comparable performance to CNNs. The optimisation of parameters contributed to better segmentations, resulting in an average increase of 1.7% and 5.0% in liver and tumour segmentation Dice coefficients, respectively. In conclusion, the findings of this study indicate that hybrid CNN/transformer architectures may serve as a practical substitute for CNNs even in small datasets. This underscores the significance of hyperparameter optimisation.

Source-free unsupervised adaptive segmentation for knee joint MRI

Knee osteoarthritis is a prevalent disease worldwide. The automatic segmentation of knee tissues in magnetic resonance (MR) images has important clinical utility in assessing knee osteoarthritis. Deep learning-based methods show great potential in this application, but they often require a large amount of labeled training data, which is challenging and expensive to acquire. Unsupervised domain adaptation that transfers the learned knowledge from a source labeled dataset to a target unlabeled dataset can be used to address this problem. However, in medical scenarios, domain adaption techniques are often limited by access to the source data due to concerns about patient privacy. In this work, we proposed a novel and effective source-free unsupervised domain adaptation method for knee joint multi-tissue segmentation that does not require a source dataset. The proposed framework is split into two stages. In the first stage, matching batch normalization statistics guides the first segmentation network to realize model adaptation and is combined with augmented entropy minimization to obtain pseudo segmentation labels for the target MR images. The pseudo labels generated by the first segmentation network are then refined using a voting strategy to supervise the training of the models in the second stage. In the second stage, uncertainty-aware cross pseudo supervision is used to further boost the performance of the desired segmentation network, which comprises an encoder and a primary decoder. Our experiments demonstrated that the proposed method outperforms the current state-of-the-art source-free unsupervised domain adaptation techniques for segmenting knee tissues in MR images. Our study also demonstrated that the proposed approach is effective in reverse-directional adaptation.

Deep learning applications in MRI for brain tumor detection and image segmentation

Deep learning holds great potential in the field of MRI applications. By leveraging its advanced algorithms and neural networks, it can effectively analyze and interpret intricate patterns in medical images, aiding in precise disease detection, segmentation, and classification. Integrating deep learning techniques with MRI technology is expected to revolutionize radiology practice, facilitating enhanced diagnostic accuracy and customized treatment strategies, ultimately leading to improved patient outcomes. This article provides an overview of of the latest advancements in deep learning techniques applied to magnetic resonance imaging, specifically focusing on brain tumor detection and segmentation. The study examines eight different deep learning methods, including a multi-scale convolutional neural network, U-Net-based fully convolutional networks, cascaded anisotropic convolutional neural networks, missing modality-based tumor segmentation, Hough-CNN for deep brain region segmentation, k-Space deep learning for accelerated MRI, Multi-level Kronecker Convolutional Neural Network, and a heuristic approach for clinical brain tumor segmentation. Each method is analyzed, highlighting its specific techniques, advantages, and limitations. The comparative performance of these methods in terms of accuracy and efficiency, addressing key factors such as computational requirements, training time, and robustness, was discussed in this article. By assessing the merits and limitations of different approaches, this review seeks to offer valuable perspectives on effective utilization of deep learning techniques in clinical MRI settings for detecting and delineating brain tumors.