Hausdorff Distance Research Articles

Autocontouring of primary lung lesions and nodal disease for radiotherapy based only on computed tomography images

Background and purposeIn many clinics, positron-emission tomography is unavailable and clinician time extremely limited. Here we describe a deep-learning model for autocontouring gross disease for patients undergoing palliative radiotherapy for primary lung lesions and/or hilar/mediastinal nodal disease, based only on computed tomography (CT) images. Materials and methodsAn autocontouring model (nnU-Net) was trained to contour gross disease in 379 cases (352 training, 27 test); 11 further test cases from an external centre were also included. Anchor-point-based post-processing was applied to remove extraneous autocontoured regions. The autocontours were evaluated quantitatively in terms of volume similarity (Dice similarity coefficient [DSC], surface Dice coefficient, 95th percentile Hausdorff distance [HD95], and mean surface distance), and scored for usability by two consultant oncologists. The magnitude of treatment margin needed to account for geometric discrepancies was also assessed. ResultsThe anchor point process successfully removed all erroneous regions from the autocontoured disease, and identified two cases to be excluded from further analysis due to ‘missed’ disease. The average DSC and HD95 were 0.8 ± 0.1 and 10.5 ± 7.3 mm, respectively. A 10-mm uniform margin-distance applied to the autocontoured region was found to yield “full coverage” (sensitivity > 0.99) of the clinical contour for 64 % of cases. Ninety-seven percent of evaluated autocontours were scored by both clinicians as requiring no or minor edits. ConclusionsOur autocontouring model was shown to produce clinically usable disease outlines, based on CT alone, for approximately two-thirds of patients undergoing lung radiotherapy. Further work is necessary to improve this before clinical implementation.

Learning carotid vessel wall segmentation in black-blood MRI using sparsely sampled cross-sections from 3D data.

Atherosclerosis of the carotid artery is a major risk factor for stroke. Quantitative assessment of the carotid vessel wall can be based on cross-sections of three-dimensional (3D) black-blood magnetic resonance imaging (MRI). To increase reproducibility, a reliable automatic segmentation in these cross-sections is essential. We propose an automatic segmentation of the carotid artery in cross-sections perpendicular to the centerline to make the segmentation invariant to the image plane orientation and allow a correct assessment of the vessel wall thickness (VWT). We trained a residual U-Net on eight sparsely sampled cross-sections per carotid artery and evaluated if the model can segment areas that are not represented in the training data. We used 218 MRI datasets of 121 subjects that show hypertension and plaque in the ICA or CCA measuring in ultrasound. The model achieves a high mean Dice coefficient of 0.948/0.859 for the vessel's lumen/wall, a low mean Hausdorff distance of , and a low mean average contour distance of on the test set. The model reaches similar results for regions of the carotid artery that are not incorporated in the training set and on MRI of young, healthy subjects. The model also achieves a low median Hausdorff distance of on the 2021 Carotid Artery Vessel Wall Segmentation Challenge test set. The proposed method can reduce the effort for carotid artery vessel wall assessment. Together with human supervision, it can be used for clinical applications, as it allows a reliable measurement of the VWT for different patient demographics and MRI acquisition settings.

ScLNet: A cornea with scleral lens OCT layers segmentation dataset and new multi-task model

ObjectiveTo develop deep learning methods with high accuracy for segmenting irregular corneas and detecting the tear fluid reservoir (TFR) boundary under the scleral lens. Additionally, this study aims to provide a publicly available cornea with scleral lens OCT dataset, including manually labeled layer masks for training and validation of segmentation algorithms. This study introduces ScLNet, a dataset comprising cornea with Scleral Lens (ScL) optical coherence tomography (OCT) images with layer annotations, and a multi-task network designed to achieve rapid, accurate, automated segmentation of scleral lens with regular and irregular corneas. MethodsWe created a dataset comprising 31,360 OCT images with scleral lens annotations. The network architecture includes an encoder with multi-scale input and a context coding layer, along with two decoders for specific tasks. The primary task focuses on predicting ScL, TFR, and cornea regions, while the auxiliary task, aimed at predicting the boundaries of ScL, TFR, and cornea, enhances feature extraction for the main task. Segmentation results were compared with state-of-the-art methods and evaluated using Dice similarity coefficient (DSC), intersection over union (IoU), Matthews correlation coefficient (MCC), Precision, and Hausdorff distance (HD). ResultsScLNet achieves 98.22 % DSC, 96.50 % IoU, 98.13 % MCC, 98.35 % Precision, and 3.6840 HD (in pixels) in segmenting ScL; 97.78 % DSC, 95.66 % IoU, 97.71 % MCC, 97.70 % Precision, and 3.7838 HD (in pixels) in segmenting TFR; and 99.22 % DSC, 98.45 % IoU, 99.15 % MCC, 99.14 % Precision, and 3.5355 HD (in pixels) in segmenting cornea. The layer interfaces recognized by ScLNet closely align with expert annotations, as evidenced by high IoU scores. Boundary metrics further confirm its effectiveness. ConclusionWe constructed a dataset of corneal OCT images with ScL wearing, which includes regular and irregular cornea patients. The proposed ScLNet achieves high accuracy in extracting ScL, TFR, and corneal layer masks and boundaries from OCT images of the dataset.

Aggregate-aware model with bidirectional edge generation for medical image segmentation

Accurate segmentation of lesion areas plays an important role in medical imaging-assisted diagnosis and treatment. Accurate boundary information can help doctors develop precise surgical plans and improve patient prognosis. However, automatic segmentation methods often struggle to accurately segment edges due to the random shape, size, and location of regions of interest (ROI). This problem is compounded in medical images, where the difference in pixel intensity between foreground and background is significantly smaller than in natural images. In this study, we propose an aggregate-aware model with bidirectional edge generation (Ambeg) for medical image segmentation. To overcome the problem of blurred edges between foreground and background in medical images, we design a deep learning model via a multi-task learning strategy and obtain richer visual features to guide segmentation. Furthermore, an Edge Feature Fusion (EFF) module is developed to combine spatial correlation information of lesion edges between adjacent images for more accurate edge segmentation. Finally, we design a new evaluation metric, the Boundary DSC segmentation consistency measure, to evaluate the edge segmentation accuracy of medical image segmentation methods. We utilize dilation and erosion operations in morphological methods to construct lesion edge labels. In addition, we use expansion and erosion rates to regulate the dimensions of the edge region to assess the requirements of different diseases for edge segmentation accuracy. The proposed approach is particularly noteworthy for achieving state-of-the-art results on medical image segmentation datasets, including BraTS 2022 (MRI), BraTS 2020 (MRI) and COVID-19–20 (CT), which have different modalities of datasets. It has an impressive Hausdorff distance of 4.62 mm and a sensitivity score of 92.45 % on BraTS 2020. Compared with existing assessment methods such as Dice score, Boundary DSC segmentation consistency measure focuses on the hard-to-segment lesion edge region rather than the easy-to-segment lesion center region, which provides a more comprehensive reference for physicians to choose automatic segmentation methods. In addition, since the boundary shapes of medical images are complex and diverse, we utilize morphological methods to obtain the boundary labels to ensure the smoothness of the boundary. Moreover, the approach is easy to implement and has a low computational cost, making it an attractive option for practical medical imaging applications.

Advanced OCTA imaging segmentation: Unsupervised, non-linear retinal vessel detection using modified self-organizing maps and joint MGRF modeling

Background and Objective:This paper proposes a fully automated and unsupervised stochastic segmentation approach using two-level joint Markov-Gibbs Random Field (MGRF) to detect the vascular system from retinal Optical Coherence Tomography Angiography (OCTA) images, which is a critical step in developing Computer-Aided Diagnosis (CAD) systems for detecting retinal diseases. Methods:Using a new probabilistic model based on a Linear Combination of Discrete Gaussian (LCDG), the first level models the appearance of OCTA images and their spatially smoothed images. The parameters of the LCDG model are estimated using a modified Expectation Maximization (EM) algorithm. The second level models the maps of OCTA images, including the vascular system and other retina tissues, using MGRF with analytically estimated parameters from the input images. The proposed segmentation approach employs modified self-organizing maps as a MAP-based optimizer maximizing the joint likelihood and handles the Joint MGRF model in a new, unsupervised way. This approach deviates from traditional stochastic optimization approaches and leverages non-linear optimization to achieve more accurate segmentation results. Results:The proposed segmentation framework is evaluated quantitatively on a dataset of 204 subjects. Achieving 0.92 ± 0.03 Dice similarity coefficient, 0.69 ± 0.25 95-percentile bidirectional Hausdorff distance, and 0.93 ± 0.03 accuracy, confirms the superior performance of the proposed approach. Conclusions:The conclusions drawn from the study highlight the superior performance of the proposed unsupervised and fully automated segmentation approach in detecting the vascular system from OCTA images. This approach not only deviates from traditional methods but also achieves more accurate segmentation results, demonstrating its potential in aiding the development of CAD systems for detecting retinal diseases.

A new point cloud simplification algorithm based on V-P container constraint and normal vector angle information entropy

Abstract Most point cloud simplification algorithms use k-order neighborhood parameters, which are set by human experience; thus, the accuracy of point feature information is not high, and each point is repeatedly calculated simultaneously. The proposed method avoids this problem. The first ordinal point of the original point cloud file was used as the starting point, and the same spatial domain was then described. The design method filters out points located in the same spatial domain and stores them in the same V-P container. The normal vector angle information entropy was calculated for each point in each container. Points with information entropy values that met the threshold requirements were extracted and stored as simplified points and new seed points. In the second operation, a point from the seed point set was selected as the starting point for the operation. The same process was repeated as the first operation. After the operation, the point from the seed point set was deleted. This process was repeated until the seed point set was empty and the algorithm ended. The simplified point set thus obtained was the simplified result. Five experimental datasets were selected and compared using the five advanced methods. The results indicate that the proposed method maintains a simplification rate of over 82% and reduces the maximum error, average error, and Hausdorff distance by 0.1099, 0.074, and 0.0062 (the highest values among the five datasets), respectively. This method has superior performance for single object and multi object point cloud sets, particularly as a reference for the study of simplified algorithms for more complex, multi object and ultra-large point cloud sets obtained using terrestrial laser scanning and mobile laser scanning.

A second-order in time, BGN-based parametric finite element method for geometric flows of curves

Over the last two decades, the field of geometric curve evolutions has attracted significant attention from scientific computing. One of the most popular numerical methods for solving geometric flows is the so-called BGN scheme, which was proposed by Barrett et al. (2007) [8], due to its favorable properties (e.g., its computational efficiency and the good mesh property). However, the BGN scheme is limited to first-order accuracy in time, and how to develop a higher-order numerical scheme is challenging. In this paper, we propose a fully discrete, temporal second-order parametric finite element method, which integrates with two different mesh regularization techniques, for solving geometric flows of curves. The scheme is constructed based on the BGN formulation and a semi-implicit Crank-Nicolson leap-frog time stepping discretization as well as a linear finite element approximation in space. More importantly, we point out that the shape metrics, such as manifold distance and Hausdorff distance, instead of function norms, should be employed to measure numerical errors. Extensive numerical experiments demonstrate that the proposed BGN-based scheme is second-order accurate in time in terms of shape metrics. Moreover, by employing the classical BGN scheme as mesh regularization techniques, our proposed second-order schemes exhibit good properties with respect to the mesh distribution. In addition, an unconditional interlaced energy stability property is obtained for one of the mesh regularization techniques.

Deep Learning Algorithms for Bladder Cancer Segmentation on Multi-Parametric MRI.

Bladder cancer (BC) segmentation on MRI images is the first step to determining the presence of muscular invasion. This study aimed to assess the tumor segmentation performance of three deep learning (DL) models on multi-parametric MRI (mp-MRI) images. We studied 53 patients with bladder cancer. Bladder tumors were segmented on each slice of T2-weighted (T2WI), diffusion-weighted imaging/apparent diffusion coefficient (DWI/ADC), and T1-weighted contrast-enhanced (T1WI) images acquired at a 3Tesla MRI scanner. We trained Unet, MAnet, and PSPnet using three loss functions: cross-entropy (CE), dice similarity coefficient loss (DSC), and focal loss (FL). We evaluated the model performances using DSC, Hausdorff distance (HD), and expected calibration error (ECE). The MAnet algorithm with the CE+DSC loss function gave the highest DSC values on the ADC, T2WI, and T1WI images. PSPnet with CE+DSC obtained the smallest HDs on the ADC, T2WI, and T1WI images. The segmentation accuracy overall was better on the ADC and T1WI than on the T2WI. The ECEs were the smallest for PSPnet with FL on the ADC images, while they were the smallest for MAnet with CE+DSC on the T2WI and T1WI. Compared to Unet, MAnet and PSPnet with a hybrid CE+DSC loss function displayed better performances in BC segmentation depending on the choice of the evaluation metric.

Optimizing Performance of Transformer-based Models for Fetal Brain MR Image Segmentation.

Purpose To test the performance of a transformer-based model when manipulating pretraining weights, dataset size, and input size and comparing the best model with the reference standard and state-of-the-art models for a resting-state functional (rs-fMRI) fetal brain extraction task. Materials and Methods An internal retrospective dataset (172 fetuses, 519 images; collected 2018-2022) was used to investigate influence of dataset size, pretraining approaches, and image input size on Swin-U-Net transformer (UNETR) and UNETR models. The internal and external (131 fetuses, 561 images) datasets were used to cross-validate and to assess generalization capability of the best model versus state-of-the-art models on different scanner types and number of gestational weeks (GWs). The Dice similarity coefficient (DSC) and the balanced average Hausdorff distance (BAHD) were used as segmentation performance metrics. Generalized equation estimation multifactorial models were used to assess significant model and interaction effects of interest. Results The Swin-UNETR model was not affected by the pretraining approach and dataset size and performed best with the mean dataset image size, with a mean DSC of 0.92 and BAHD of 0.097. Swin-UNETR was not affected by scanner type. Generalization results on the internal dataset showed that Swin-UNETR had lower performance compared with the reference standard models and comparable performance on the external dataset. Cross-validation on internal and external test sets demonstrated better and comparable performance of Swin-UNETR versus convolutional neural network architectures during the late-fetal period (GWs > 25) but lower performance during the midfetal period (GWs ≤ 25). Conclusion Swin-UNTER showed flexibility in dealing with smaller datasets, regardless of pretraining approaches. For fetal brain extraction from rs-fMR images, Swin-UNTER showed comparable performance with that of reference standard models during the late-fetal period and lower performance during the early GW period. Keywords: Transformers, CNN, Medical Imaging Segmentation, MRI, Dataset Size, Input Size, Transfer Learning Supplemental material is available for this article. © RSNA, 2024.

Ischemic stroke infarct segmentation model based on depthwise separable convolution for multimodal magnetic resonance imaging

Magnetic resonance imaging (MRI) plays a crucial role in the diagnosis of ischemic stroke. Accurate segmentation of the infarct is of great significance for selecting intervention treatment methods and evaluating the prognosis of patients. To address the issue of poor segmentation accuracy of existing methods for multiscale stroke lesions, a novel encoder-decoder architecture network based on depthwise separable convolution is proposed. Firstly, this network replaces the convolutional layer modules of the U-Net with redesigned depthwise separable convolution modules. Secondly, an modified Atrous spatial pyramid pooling (MASPP) is introduced to enlarge the receptive field and enhance the extraction of multiscale features. Thirdly, an attention gate (AG) structure is incorporated at the skip connections of the network to further enhance the segmentation accuracy of multiscale targets. Finally, Experimental evaluations are conducted using the ischemic stroke lesion segmentation 2022 challenge (ISLES2022) dataset. The proposed algorithm in this paper achieves Dice similarity coefficient (DSC), Hausdorff distance (HD), sensitivity (SEN), and precision (PRE) scores of 0.816 5, 3.668 1, 0.889 2, and 0.894 6, respectively, outperforming other mainstream segmentation algorithms. The experimental results demonstrate that the method in this paper effectively improves the segmentation of infarct lesions, and is expected to provide a reliable support for clinical diagnosis and treatment.

Multi-scale medical image segmentation based on pixel encoding and spatial attention mechanism

In response to the issues of single-scale information loss and large model parameter size during the sampling process in U-Net and its variants for medical image segmentation, this paper proposes a multi-scale medical image segmentation method based on pixel encoding and spatial attention. Firstly, by redesigning the input strategy of the Transformer structure, a pixel encoding module is introduced to enable the model to extract global semantic information from multi-scale image features, obtaining richer feature information. Additionally, deformable convolutions are incorporated into the Transformer module to accelerate convergence speed and improve module performance. Secondly, a spatial attention module with residual connections is introduced to allow the model to focus on the foreground information of the fused feature maps. Finally, through ablation experiments, the network is lightweighted to enhance segmentation accuracy and accelerate model convergence. The proposed algorithm achieves satisfactory results on the Synapse dataset, an official public dataset for multi-organ segmentation provided by the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), with Dice similarity coefficient (DSC) and 95% Hausdorff distance (HD95) scores of 77.65 and 18.34, respectively. The experimental results demonstrate that the proposed algorithm can enhance multi-organ segmentation performance, potentially filling the gap in multi-scale medical image segmentation algorithms, and providing assistance for professional physicians in diagnosis.

Comprehensive benchmarking of CNN-based tumor segmentation methods using multimodal MRI data

Magnetic resonance imaging (MRI) has become an essential and a frontline technique in the detection of brain tumor. However, segmenting tumors manually from scans is laborious and time-consuming. This has led to an increasing trend towards fully automated methods for precise tumor segmentation in MRI scans. Accurate tumor segmentation is crucial for improved diagnosis, treatment, and prognosis. This study benchmarks and evaluates four widely used CNN-based methods for brain tumor segmentation CaPTk, 2DVNet, EnsembleUNets, and ResNet50. Using 1251 multimodal MRI scans from the BraTS2021 dataset, we compared the performance of these methods against a reference dataset of segmented images assisted by radiologists. This comparison was conducted using segmented images directly and further by radiomic features extracted from the segmented images using pyRadiomics. Performance was assessed using the Dice Similarity Coefficient (DSC) and Hausdorff Distance (HD). EnsembleUNets excelled, achieving a DSC of 0.93 and an HD of 18, outperforming the other methods. Further comparative analysis of radiomic features confirmed EnsembleUNets as the most precise segmentation method, surpassing other methods. EnsembleUNets recorded a Concordance Correlation Coefficient (CCC) of 0.79, a Total Deviation Index (TDI) of 1.14, and a Root Mean Square Error (RMSE) of 0.53, underscoring its superior performance. We also performed validation on an independent dataset of 611 samples (UPENN-GBM), which further supported the accuracy of EnsembleUNets, with a DSC of 0.85 and an HD of 17.5. These findings provide valuable insight into the efficacy of EnsembleUNets, supporting informed decisions for accurate brain tumor segmentation.

Interval-valued (p,q,r)-spherical fuzzy sets and their applications in MCGDM and MCDM based on TOPSIS method and aggregation operators

The (p,q,r)-Spherical fuzzy set is a useful generalization of other fuzzy structures such as fuzzy, intuitionistic fuzzy, Pythagorean fuzzy, picture fuzzy, q-rung orthopair fuzzy, spherical fuzzy, and T-spherical fuzzy sets. In this paper, we define concept of interval-valued (p,q,r)-spherical fuzzy set (IVpqr-SFS) and Set-theoretical and arithmetic operations of them. Two aggregation operators named interval-valued (p,q,r)-spherical fuzzy weighted arithmetic (IVpqr-SFWAA) and interval-valued (p,q,r)-spherical fuzzy weighted geometric (IVpqr-SFWGA) aggregation operators are introduced for IVpqr-SF numbers (IVpqr-SFN) and their properties were examined. The proposed operators are explained with examples. Furthermore, the score and accuracy functions of an IVpqr-SFN are introduced. We also define distance measures between two same types IVpqr-SFSs based on Hamming, Euclidean, and Hausdorff distance measures. Moreover, a multi-criteria group decision-making (MCGDM) method is developed based on the TOPSIS method by using the proposed distance measures and score function, and a numerical example is given to show the process of the proposed MCGDM method. In addition, a multi-criteria decision-making (MCDM) method is developed using the proposed IVpqr-SFWAA, IVpqr-SFWGA operators, and an illustrative example is given to show the process of the proposed MCDM method. Finally, a comparative analysis is presented between the proposed methods and the existing methods.

Radiomics as a measure superior to common similarity metrics for tumor segmentation performance evaluation.

To propose radiomics features as a superior measure for evaluating the segmentation ability of physicians and auto-segmentation tools and to compare its performance with the most commonly used metrics: Dice similarity coefficient (DSC), surface Dice similarity coefficient (sDSC), and Hausdorff distance (HD). The data of 10 lung cancer patients' CT images with nine tumor segmentations per tumor were downloaded from the RIDER (Reference Database to Evaluate Response) database. Radiomics features of 90 segmented tumors were extracted using the PyRadiomics program. The intraclass correlation coefficient (ICC) of radiomics features were used to evaluate the segmentation similarity and compare their performance with DSC, sDSC, and HD. We calculated one ICC per radiomics feature and per tumor for nine segmentations and 36 ICCs per radiomics feature for 36 pairs of nine segmentations. Meanwhile, there were 360 DSC, sDSC, and HD values calculated for 36 pairs for 10 tumors. The ICC of radiomics features exhibited greater sensitivity to segmentation changes than DSC and sDSC. The ICCs of the wavelet-LLL first order Maximum, wavelet-LLL glcm MCC, wavelet-LLL glcm Cluster Shade features ranged from 0.130 to 0.997, 0.033 to 0.978, and 0.160 to 0.998, respectively. On the other hand, all DSC and sDSC were larger than 0.778 and 0.700, respectively, while HD varied from 0 to 1.9mm. The results indicated that the radiomics features could capture subtle variations in tumor segmentation characteristics, which could not be easily detected by DSC and sDSC. This study demonstrates the superiority of radiomics features with ICC as a measure for evaluating a physician's tumor segmentation ability and the performance of auto-segmentation tools. Radiomics features offer a more sensitive and comprehensive evaluation, providing valuable insights into tumor characteristics. Therefore, the new metrics can be used to evaluate new auto-segmentation methods and enhance trainees' segmentation skills in medical training and education.

Automatic Segmentation of Type A Aortic Dissection on Computed Tomography Images Using Deep Learning Approach.

Type A aortic dissection (TAAD) is a life-threatening aortic disease. The tear involves the ascending aorta and progresses into the separation of the layers of the aortic wall and the occurrence of a false lumen. Accurate segmentation of TAAD could provide assistance for disease assessment and guidance for clinical treatment. This study applied nnU-Net, a state-of-the-art biomedical segmentation network architecture, to segment contrast-enhanced CT images and quantify the morphological features for TAAD. CT datasets were acquired from 24 patients with TAAD. Manual segmentation and annotation of the CT images was used as the ground-truth. Two-dimensional (2D) nnU-Net and three-dimensional (3D) nnU-Net architectures with Dice- and cross entropy-based loss functions were utilized to segment the true lumen (TL), false lumen (FL), and intimal flap on the images. Four-fold cross validation was performed to evaluate the performance of the two nnU-Net architectures. Six metrics, including accuracy, precision, recall, Intersection of Union, Dice similarity coefficient (DSC), and Hausdorff distance, were calculated to evaluate the performance of the 2D and 3D nnU-Net algorithms in TAAD datasets. Aortic morphological features from both 2D and 3D nnU-Net algorithms were quantified based on the segmented results and compared. Overall, 3D nnU-Net architectures had better performance in TAAD CT datasets, with TL and FL segmentation accuracy up to 99.9%. The DSCs of TLs and FLs based on the 3D nnU-Net were 88.42% and 87.10%. For the aortic TL and FL diameters, the FL area calculated from the segmentation results of the 3D nnU-Net architecture had smaller relative errors (3.89-6.80%), compared to the 2D nnU-Net architecture (relative errors: 4.35-9.48%). The nnU-Net architectures may serve as a basis for automatic segmentation and quantification of TAAD, which could aid in rapid diagnosis, surgical planning, and subsequent biomechanical simulation of the aorta.

HaN-Seg: The head and neck organ-at-risk CT and MR segmentation challenge

Background and purposeTo promote the development of auto-segmentation methods for head and neck (HaN) radiation treatment (RT) planning that exploit the information of computed tomography (CT) and magnetic resonance (MR) imaging modalities, we organized HaN-Seg: The Head and Neck Organ-at-Risk CT and MR Segmentation Challenge. Materials and methodsThe challenge task was to automatically segment 30 organs-at-risk (OARs) of the HaN region in 14 withheld test cases given the availability of 42 publicly available training cases. Each case consisted of one contrast-enhanced CT and one T1-weighted MR image of the HaN region of the same patient, with up to 30 corresponding reference OAR delineation masks. The performance was evaluated in terms of the Dice similarity coefficient (DSC) and 95-percentile Hausdorff distance (HD95), and statistical ranking was applied for each metric by pairwise comparison of the submitted methods using the Wilcoxon signed-rank test. ResultsWhile 23 teams registered for the challenge, only seven submitted their methods for the final phase. The top-performing team achieved a DSC of 76.9 % and a HD95 of 3.5 mm. All participating teams utilized architectures based on U-Net, with the winning team leveraging rigid MR to CT registration combined with network entry-level concatenation of both modalities. ConclusionThis challenge simulated a real-world clinical scenario by providing non-registered MR and CT images with varying fields-of-view and voxel sizes. Remarkably, the top-performing teams achieved segmentation performance surpassing the inter-observer agreement on the same dataset. These results set a benchmark for future research on this publicly available dataset and on paired multi-modal image segmentation in general.

Soap bubbles and convex cones: optimal quantitative rigidity

We consider a class of recent rigidity results in a convex cone Σ ⊆ R N \Sigma \subseteq \mathbb {R}^N . These include overdetermined Serrin-type problems for a mixed boundary value problem relative to Σ \Sigma , Alexandrov’s soap bubble-type results relative to Σ \Sigma , and Heintze-Karcher’s inequality relative to Σ \Sigma . Each rigidity result is obtained here by means of a single integral identity and holds true under weak integral overdeterminations in possibly non-smooth cones. Optimal quantitative stability estimates are obtained in terms of an L 2 L^2 -pseudodistance. In particular, the optimal stability estimate for Heintze-Karcher’s inequality is new even in the classical case Σ = R N \Sigma = \mathbb {R}^N . Stability bounds in terms of the Hausdorff distance are also provided. Several new results are established and exploited, including a new Poincaré-type inequality for vector fields whose normal components vanish on a portion of the boundary and an explicit (possibly weighted) trace theory – relative to the cone Σ \Sigma – for harmonic functions satisfying a homogeneous Neumann condition on the portion of the boundary contained in ∂ Σ \partial \Sigma . We also introduce new notions of uniform interior and exterior sphere conditions relative to the cone Σ ⊆ R N \Sigma \subseteq \mathbb {R}^N , which allow to obtain (via barrier arguments) uniform lower and upper bounds for the gradient in the mixed boundary value-setting. In the particular case Σ = R N \Sigma =\mathbb {R}^N , these conditions return the classical uniform interior and exterior sphere conditions (together with the associated classical gradient bounds of the Dirichlet setting).

Comparison of the use of a clinically implemented deep learning segmentation model with the simulated study setting for breast cancer patients receiving radiotherapy.

Deep learning (DL) models for auto-segmentation in radiotherapy have been extensively studied in retrospective and pilot settings. However, these studies might not reflect the clinical setting. This study compares the use of a clinically implemented in-house trained DL segmentation model for breast cancer to a previously performed pilot study to assess possible differences in performance or acceptability. Sixty patients with whole breast radiotherapy, with or without an indication for locoregional radiotherapy were included. Structures were qualitatively scored by radiotherapy technologists and radiation oncologists. Quantitative evaluation was performed using dice-similarity coefficient (DSC), 95th percentile of Hausdorff Distance (95%HD) and surface DSC (sDSC), and time needed for generating, checking, and correcting structures was measured. Ninety-three percent of all contours in clinic were scored as clinically acceptable or usable as a starting point, comparable to 92% achieved in the pilot study. Compared to the pilot study, no significant changes in time reduction were achieved for organs at risks (OARs). For target volumes, significantly more time was needed compared to the pilot study for patients including lymph node levels 1-4, although time reduction was still 33% compared to manual segmentation. Almost all contours have better DSC and 95%HD than inter-observer variations. Only CTVn4 scored worse for both metrics, and the thyroid had a higher 95%HD value. The use of the DL model in clinical practice is comparable to the pilot study, showing high acceptability rates and time reduction.

Multi-modal segmentation with missing image data for automatic delineation of gross tumor volumes in head and neck cancers.

Head and neck (HN) gross tumor volume (GTV) auto-segmentation is challenging due to the morphological complexity and low image contrast of targets. Multi-modality images, including computed tomography (CT) and positron emission tomography (PET), are used in the routine clinic to assist radiation oncologists for accurate GTV delineation. However, the availability of PET imaging may not always be guaranteed. To develop a deep learning segmentation framework for automated GTV delineation of HN cancers using a combination of PET/CT images, while addressing the challenge of missing PET data. Two datasets were included for this study: Dataset I: 524 (training) and 359 (testing) oropharyngeal cancer patients from different institutions with their PET/CT pairs provided by the HECKTOR Challenge; Dataset II: 90 HN patients(testing) from a local institution with their planning CT, PET/CT pairs. To handle potentially missing PET images, a model training strategy named the "Blank Channel" method was implemented. To simulate the absence of a PET image, a blank array with the same dimensions as the CT image was generated to meet the dual-channel input requirement of the deep learning model. During the model training process, the model was randomly presented with either a real PET/CT pair or a blank/CT pair. This allowed the model to learn the relationship between the CT image and the corresponding GTV delineation based on available modalities. As a result, our model had the ability to handle flexible inputs during prediction, making it suitable for cases where PET images are missing. To evaluate the performance of our proposed model, we trained it using training patients from Dataset I and tested it with Dataset II. We compared our model (Model 1) with two other models which were trained for specific modality segmentations: Model 2 trained with only CT images, and Model 3 trained with real PET/CT pairs. The performance of the models was evaluated using quantitative metrics, including Dice similarity coefficient (DSC), mean surface distance (MSD), and 95% Hausdorff Distance (HD95). In addition, we evaluated our Model 1 and Model 3 using the 359 test cases in Dataset I. Our proposed model(Model 1) achieved promising results for GTV auto-segmentation using PET/CT images, with the flexibility of missing PET images. Specifically, when assessed with only CT images in Dataset II, Model 1 achieved DSC of 0.56±0.16, MSD of 3.4±2.1mm, and HD95 of 13.9±7.6mm. When the PET images were included, the performance of our model was improved to DSC of 0.62±0.14, MSD of 2.8±1.7mm, and HD95 of 10.5±6.5mm. These results are comparable to those achieved by Model 2 and Model 3, illustrating Model 1's effectiveness in utilizing flexible input modalities. Further analysis using the test dataset from Dataset I showed that Model 1 achieved an average DSC of 0.77, surpassing the overall average DSC of 0.72 among all participants in the HECKTOR Challenge. We successfully refined a multi-modal segmentation tool for accurate GTV delineation for HN cancer. Our method addressed the issue of missing PET images by allowing flexible data input, thereby providing a practical solution for clinical settings where access to PET imaging may be limited.

Subject-Specific Probability Maps of Scalp, Skull and Cerebrospinal Fluid for Cranial Bones Segmentation in Neonatal Cerebral MRIs

ObjectivesSegmentation of cranial bones in magnetic resonance images (MRIs) is a challenging and indispensable task to study neonatal brain development and injury. This paper presents a new approach for creating subject-specific probability maps of the scalp, skull and cerebrospinal fluid (CSF) from retrospective bimodal (MR and CT) images acquired from neonates in the gestational age range of 39 to 42 weeks. These maps are subsequently employed for the segmentation of cranial bones in cerebral MRIs from neonates in the same age range. Material and methodsRetrospective MR and CT of neonates with normal head in the gestational age range of 39-42 weeks were preprocessed, segmented semi-automatically and employed as atlas data. For an input MR image acquired from a subject under study, a preprocessing stage and three main processing blocks were performed: First, subject-specific head and intracranial templates and CSF probability map were created using retrospective MR atlas data. Second, the CT atlas data were coregistered to MR templates and the resulted deformation matrices were fed to the next block to create subject-specific scalp and skull probability maps. Finally, some novel performance measures were presented to evaluate the performance of subject-specific CSF, scalp and skull probability maps for skull and intracranial segmentation in neonatal MRIs. ResultsThe subject-specific probability maps were employed for brain tissue extraction and compared with two public methods such as Brain Extraction Tool (BET) and Brain Surface Extractor (BSE). They were also applied for cranial bone extraction. Then, the similarity in shape between the frontal and occipital sutures (which had been reconstructed from segmented cranial bones) and the ground truth landmarks was evaluated. For this purpose, modified versions of the Dice similarity coefficient (DSC) were used. Finally, a retrospective bimodal (MR-CT) data acquired from a neonate within a short time interval was used for evaluation. After co-alignment of the two images, the DSC and modified Hausdorff distance (MHD) were used to compare the similarity of cranial bones in the MR and CT images. ConclusionSignificant improvements were achieved compared to conventional methods which rely solely on MR image intensities. These advancements hold promise for enhancing neurodevelopmental studies in neonates. The algorithm for creating subject-specific atlases is publicly accessible through a graphical user interface at medvispy.ee.kntu.ac.ir.