Symmetric Surface Research Articles

Abstract P03: Identifying sources of variation in manual segmentation of hippocampus on from magnetic resonance images (MRI)

Abstract The purpose of the present study is to evaluate inter-operator and intra-operator variations in the manual segmentation of the hippocampus from high resolution T1-weighted magnetic resonance (MR) images. The hippocampus was segmented manually in MR images of 118 epileptic and 25 non-epileptic patients (65 males, 78 females; median age of 36 years, mean age of 39 years) by three operators (M1, M2, M3) and three automated methods (FreeSurfer, LocalInfo, ABSS). To determine how much the manual segmentations performed by one operator differ from those of another operator, inter-operator variability was evaluated. To determine how much the manual segmentations done by an operator vary over time, i.e., to assess intra-operator variability, manual segmentations from each operator were compared to the automated segmentations. To this end, rational absolute value degree (RAVD), volume asymmetry, Dice coefficient, precision, similarity, specificity, accuracy, negative predictive value (NPV), Hausdorff distance, root mean square (RMS), average symmetric surface distance (ASSD), and mean distance were calculated. The segmentation results of one of the operators were considered as the ground truth for the evaluation of the segmentation results of the other operators and the automatic segmentation methods. Hausdorff distance and precision were different when using automated techniques as the test segmentation and the M3 segmentation as the ground truth, rather than M1 and M2 segmentations. The standard deviation of performance measures tended to be higher when using operator M3 as the ground truth and either operator M1 or M2 as the test segmentation. Variation in performance measures when using M3 as the ground truth is indicative of inter-operator variation. When comparing performance measures generated by automated versus manual techniques, standard deviations were larger when using operator M2 as the ground truth than when using operator M1. This suggests that operator M2 exhibited a larger intra-operator variation than operator M1. Among the automatic segmentation methods, ABSS was the most effective method in many regards (RAVD Dice coefficient, similarity, specificity, accuracy, NPV, Hausdorff distance, RMS, ASSD) while FreeSurfer and LocalInfo were more effective for the precision, mean distance, and lateralization of epileptogenicity. Inter-operator error was likely due to the temporal separation of the segmentations and thus, it may be reduced by having all operators working in the laboratory simultaneously and undergoing the same training, although some inter-operator variability may be unavoidable. Intra-operator variation can likely be reduced with further training and supervision of the operators by a neuroradiologist with expertise in hippocampus anatomy. Future automated segmentation techniques may incorporate elements of both atlas-based (FreeSurfer and LocalInfo) and neural-network-based (ABSS) segmentation techniques for optimal performance. Citation Format: Benjamin Huber, Esmaeil Davoodi-Bojd, Hamid Soltanian-Zadeh. Identifying sources of variation in manual segmentation of hippocampus on from magnetic resonance images (MRI) [abstract]. In: Proceedings of the AACR Virtual Meeting: COVID-19 and Cancer; 2021 Feb 3-5. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(6_Suppl):Abstract nr P03.

Automated Segmentation of Infarct Lesions in T1-Weighted MRI Scans Using Variational Mode Decomposition and Deep Learning.

Automated segmentation methods are critical for early detection, prompt actions, and immediate treatments in reducing disability and death risks of brain infarction. This paper aims to develop a fully automated method to segment the infarct lesions from T1-weighted brain scans. As a key novelty, the proposed method combines variational mode decomposition and deep learning-based segmentation to take advantages of both methods and provide better results. There are three main technical contributions in this paper. First, variational mode decomposition is applied as a pre-processing to discriminate the infarct lesions from unwanted non-infarct tissues. Second, overlapped patches strategy is proposed to reduce the workload of the deep-learning-based segmentation task. Finally, a three-dimensional U-Net model is developed to perform patch-wise segmentation of infarct lesions. A total of 239 brain scans from a public dataset is utilized to develop and evaluate the proposed method. Empirical results reveal that the proposed automated segmentation can provide promising performances with an average dice similarity coefficient (DSC) of 0.6684, intersection over union (IoU) of 0.5022, and average symmetric surface distance (ASSD) of 0.3932, respectively.

Tropical Cyclone Frequency Under Varying SSTs in Aquaplanet Simulations

AbstractGlobal tropical cyclone (TC) frequency is investigated in a 50‐km‐resolution aquaplanet model forced by zonally symmetric sea surface temperature (SST). TC frequency per unit area is found to be proportional to the Coriolis parameter at the intertropical convergence zone (ITCZ), as defined by the latitude of maximum precipitation. As the latitude of maximum SST is shifted northward from the equator, the precipitation maximum moves northward and TC frequency increases. When the SST maximum is shifted northward past 25°N, the precipitation maximum remains between 15°N and 20°N, and TC frequency per unit area is approximately constant. When applied to observed precipitation and SST data, the same scaling captures a substantial fraction of observed TCs. Results suggest that future changes in TC activity will be modulated by changes in the large‐scale circulation, and in particular that the ITCZ location is an important determinant of the number of TCs.

Idealized Aquaplanet Simulations of Tropical Cyclone Activity: Significance of Temperature Gradients, Hadley Circulation, and Zonal Asymmetry

AbstractEarlier studies have proposed many semiempirical relations between climate and tropical cyclone (TC) activity. To explore these relations, this study conducts idealized aquaplanet experiments using both symmetric and asymmetric sea surface temperature (SST) forcings. With zonally symmetric SST forcings that have a maximum at 10°N, reducing meridional SST gradients around an Earth-like reference state leads to a weakening and southward displacement of the intertropical convergence zone. With nearly flat meridional gradients, warm-hemisphere TC numbers increase by nearly 100 times due particularly to elevated high-latitude TC activity. Reduced meridional SST gradients contribute to a poleward expansion of the tropics, which is associated with a poleward migration of the latitudes where TCs form or reach their lifetime maximum intensity. However, these changes cannot be simply attributed to the poleward expansion of Hadley circulation. Introducing zonally asymmetric SST forcings tends to decrease the global TC number. Regional SST warming—prescribed with or without SST cooling at other longitudes—affects local TC activity but does not necessarily increase TC genesis. While regional warming generally suppresses TC activity in remote regions with relatively cold SSTs, one experiment shows a surprisingly large increase of TC genesis. This increase of TC genesis over relatively cold SSTs is related to local tropospheric cooling that reduces static stability near 15°N and vertical wind shear around 25°N. Modeling results are discussed with scaling analyses and have implications for the application of the “convective quasi-equilibrium and weak temperature gradient” framework.

New quality weight used for phase unwrapping in color fringe reflection method

The fringe -reflection technology has been widely applied in the fields of industrial inspection and measurements. In the present study, three-step phase-shifting is developed using the color fringe-reflection approach to inspect a work-piece surface with rotational symmetry characteristics, and an improved experiment is proposed. In addition, by combining Robert gradient with fringe modulation, a new quality weight, referred to as ‘Modulation-Robert gradient variance’ (MRGV), is proposed for quality-guiding the phase-unwrapping (PU) algorithm. In comparison to the traditional quality weights, MRGV presented higher noise immunity and lower PU error in the root mean square (RMS). In addition, in terms of defect detection, MRGV performed better than the other weights. Finally, after MRGV-guided PU conduction, the proposed experiment method is used in combination with an improved Southwell integration algorithm to realize the 3D reconstruction. The reconstruction error is approximately 3.09% lower than that presented by certain other methods. Furthermore, the results indicated that the combination of the proposed method and the new quality weight could effectively detect the defects and offer an accurate measurement of a rotationally symmetrical surface to a certain extent.

The International Workshop on Osteoarthritis Imaging Knee MRI Segmentation Challenge: A Multi-Institute Evaluation and Analysis Framework on a Standardized Dataset.

To organize a multi-institute knee MRI segmentation challenge for characterizing the semantic and clinical efficacy of automatic segmentation methods relevant for monitoring osteoarthritis progression. A dataset partition consisting of three-dimensional knee MRI from 88 retrospective patients at two time points (baseline and 1-year follow-up) with ground truth articular (femoral, tibial, and patellar) cartilage and meniscus segmentations was standardized. Challenge submissions and a majority-vote ensemble were evaluated against ground truth segmentations using Dice score, average symmetric surface distance, volumetric overlap error, and coefficient of variation on a holdout test set. Similarities in automated segmentations were measured using pairwise Dice coefficient correlations. Articular cartilage thickness was computed longitudinally and with scans. Correlation between thickness error and segmentation metrics was measured using the Pearson correlation coefficient. Two empirical upper bounds for ensemble performance were computed using combinations of model outputs that consolidated true positives and true negatives. Six teams (T 1-T 6) submitted entries for the challenge. No differences were observed across any segmentation metrics for any tissues (P = .99) among the four top-performing networks (T 2, T 3, T 4, T 6). Dice coefficient correlations between network pairs were high (> 0.85). Per-scan thickness errors were negligible among networks T 1-T 4 (P = .99), and longitudinal changes showed minimal bias (< 0.03 mm). Low correlations (ρ < 0.41) were observed between segmentation metrics and thickness error. The majority-vote ensemble was comparable to top-performing networks (P = .99). Empirical upper-bound performances were similar for both combinations (P = .99). Diverse networks learned to segment the knee similarly, where high segmentation accuracy did not correlate with cartilage thickness accuracy and voting ensembles did not exceed individual network performance.See also the commentary by Elhalawani and Mak in this issue.Keywords: Cartilage, Knee, MR-Imaging, Segmentation © RSNA, 2020Supplemental material is available for this article.

The influence of a pitchfork bifurcation of the critical points of a symmetric caldera potential energy surface on dynamical matching

Many organic chemical reactions are governed by potential energy surfaces that have a region with the topographical features of a caldera. If the caldera has a symmetry then trajectories transiting the caldera region are observed to exhibit a phenomenon that is referred to as dynamical matching. Dynamical matching is a constraint that restricts the way in which a trajectory can exit the caldera based solely on how it enters the caldera. In this paper we show that bifurcations of the critical points of the caldera potential energy surface can destroy dynamical matching even when the symmetry of the caldera is not affected by the bifurcation.

RBF Interpolation Algorithm for FTS Tool Path Generation

Freeform optics are defined as nonrotational symmetric optical surfaces in the manufacturing industry. Freeform optics are extensively applied to many areas in order to improve system performance. Fast tool servo (FTS) assisting single-point diamond turning technology has high application prospects in freeform optics machining. This paper discusses the interpolation algorithm for tool path generation of FTS through the application of a radial basis function (RBF) algorithm. For this purpose, a positive definite RBF with compact support was employed as the interpolant. The existence is mathematically proven. Numerical simulations were performed to compare the performances of the RBF algorithm and commonly used algorithms for satisfying the requirements of existence, smoothness, and accuracy. Machining experiments were also conducted to validate the applicability of the algorithm. The simulation results showed that the RBF interpolation algorithm outperformed other algorithms in terms of smoothness. The RBF algorithm also provided the highest interpolation accuracy. Furthermore, the RBF interpolation algorithm exhibited the highest accuracy for error distribution, with large errors distributed mainly in transition areas. The machining results were also in general agreement with the simulation results although obvious practical errors were observed. Overall, RBF interpolation can provide higher accuracy and better smoothness in the tool path generation of FTS.

MI-UNet: Multi-Inputs UNet Incorporating Brain Parcellation for Stroke Lesion Segmentation From T1-Weighted Magnetic Resonance Images.

Stroke is a serious manifestation of various cerebrovascular diseases and one of the most dangerous diseases in the world today. Volume quantification and location detection of chronic stroke lesions provide vital biomarkers for stroke rehabilitation. Recently, deep learning has seen a rapid growth, with a great potential in segmenting medical images. In this work, unlike most deep learning-based segmentation methods utilizing only magnetic resonance (MR) images as the input, we propose and validate a novel stroke lesion segmentation approach named multi-inputs UNet (MI-UNet) that incorporates brain parcellation information, including gray matter (GM), white matter (WM) and lateral ventricle (LV). The brain parcellation is obtained from 3D diffeomorphic registration and is concatenated with the original MR image to form two-channel inputs to the subsequent MI-UNet. Effectiveness of the proposed pipeline is validated using a dataset consisting of 229 T1-weighted MR images. Experiments are conducted via a five-fold cross-validation. The proposed MI-UNet performed significantly better than UNet in both 2D and 3D settings. Our best results obtained by 3D MI-UNet has superior segmentation performance, as measured by the Dice score, Hausdorff distance, average symmetric surface distance, as well as precision, over other state-of-the-art methods.

Joint affine and deformable three-dimensional networks for brain MRI registration.

Volumetric medical image registration has important clinical significance. Traditional registration methods may be time-consuming when processing large volumetric data due to their iterative optimizations. In contrast, existing deep learning-based networks can obtain the registration quickly. However, most of them require independent rigid alignment before deformable registration; these two steps are often performed separately and cannot be end-to-end. We propose an end-to-end joint affine and deformable network for three-dimensional (3D) medical image registration. The proposed network combines two deformation methods; the first one is for obtaining affine alignment and the second one is a deformable subnetwork for achieving the nonrigid registration. The parameters of the two subnetworks are shared. The global and local similarity measures are used as loss functions for the two subnetworks, respectively. Moreover, an anatomical similarity loss is devised to weakly supervise the training of the whole registration network. Finally, the trained network can perform deformable registration in one forward pass. The efficacy of our network was extensively evaluated on three public brain MRI datasets including Mindboggle101, LPBA40, and IXI. Experimental results demonstrate our network consistently outperformed several state-of-the-art methods with respect to the metrics of Dice index (DSC), Hausdorff distance (HD), and average symmetric surface distance (ASSD). The proposed network provides accurate and robust volumetric registration without any pre-alignment requirement, which facilitates the end-to-end deformable registration.

High efficiency removal of single point diamond turning marks on aluminum surface by combination of ion beam sputtering and smoothing polishing.

Single point diamond turning (SPDT) is highly versatile in fabricating axially symmetric form, non-axially-symmetric form and free form surfaces. However, inevitable microstructure known as turning marks left on the surface have limited the mirror's optical performance. Based on chemical mechanical polishing (CMP) mechanism, smoothing polishing (SP) process is believed to be an effective method to remove turning marks. However, the removal efficiency is relatively low. In this paper, based on Greenwood-Williamson (GW) theory, the factors that limit removal efficiency of SP are discussed in details. Influences of process parameters (work pressure and rotational speed) are firstly discussed. With further analysis, surface spectral characteristics are identified as the inherent factor affecting further efficiency improvement. According to theoretical analysis, the removal efficiency of isotropic surface is nearly 1.8 times higher than anisotropy surface like surface with turning marks. A high efficiency turning marks removal process combining ion beam sputtering (IBS) and SP is proposed in our research. With removal depth exceeding 100 nm, the isotropic aluminum surface can be constructed by IBS so that the efficiency of SP process can be greatly improved. Though deteriorated by IBS, the surface roughness will be rapidly reduced by SP process. Finally, experiments are conducted to verify our analysis. A 3.7 nm roughness surface without turning marks is achieved by new method while direct SP can only reach roughness of 4.3 nm with evident turning marks. Experimental results show that removal efficiency nearly doubled which matches well with the theoretical analysis. Our research not only can be used as a high efficiency turning marks removal and surface quality improvement method but also can be a new method for high precision aluminum optics fabrication.

Edge-Aware Pyramidal Deformable Network for Unsupervised Registration of Brain MR Images.

Deformable image registration is of essential important for clinical diagnosis, treatment planning, and surgical navigation. However, most existing registration solutions require separate rigid alignment before deformable registration, and may not well handle the large deformation circumstances. We propose a novel edge-aware pyramidal deformable network (referred as EPReg) for unsupervised volumetric registration. Specifically, we propose to fully exploit the useful complementary information from the multi-level feature pyramids to predict multi-scale displacement fields. Such coarse-to-fine estimation facilitates the progressive refinement of the predicted registration field, which enables our network to handle large deformations between volumetric data. In addition, we integrate edge information with the original images as dual-inputs, which enhances the texture structures of image content, to impel the proposed network pay extra attention to the edge-aware information for structure alignment. The efficacy of our EPReg was extensively evaluated on three public brain MRI datasets including Mindboggle101, LPBA40, and IXI30. Experiments demonstrate our EPReg consistently outperformed several cutting-edge methods with respect to the metrics of Dice index (DSC), Hausdorff distance (HD), and average symmetric surface distance (ASSD). The proposed EPReg is a general solution for the problem of deformable volumetric registration.

Fano resonances in nanohole oligomers in a gold film

The optical properties of nanohole oligomers in a gold film have been investigated theoretically and experimentally. Helium focused ion beam milling was used to create arrays of close-packed heptamer-arranged nanoholes (HNH) in a gold film on a fused silica substrate. The holes are slightly elliptical about a nominal diameter of 100 nm and are separated by 15 nm Au regions (“bridges”) to form close-packed heptamer arrangements. Optical responses are reported as transmittance spectra, showing various Fano resonances due to different excitations on the structure: surface plasmon polaritons localized to the HNH, symmetric and asymmetric surface plasmon polaritons propagating along with the film, and surface waves associated with Wood’s anomaly. All resonances observed involve the coupling (bonding and anti-bonding) of magnetic dipoles within the holes of HNH and in some cases of magnetic hexapoles within the central hole. The sensing performance was investigated by measuring the bulk sensitivity of the resonances, anticipating application to biosensing, yielding 380 nm RIU−1 in the best case. Surface-enhanced Raman scattering measurements were also performed on HNH, where adventitious carbon was detected directly through the identification of D and G bands (spectra taken from nearby un-milled Au areas were essentially featureless).

Deep Learning-Based Automated Abdominal Organ Segmentation in the UK Biobank and German National Cohort Magnetic Resonance Imaging Studies.

The aims of this study were to train and evaluate deep learning models for automated segmentation of abdominal organs in whole-body magnetic resonance (MR) images from the UK Biobank (UKBB) and German National Cohort (GNC) MR imaging studies and to make these models available to the scientific community for analysis of these data sets. A total of 200 T1-weighted MR image data sets of healthy volunteers each from UKBB and GNC (400 data sets in total) were available in this study. Liver, spleen, left and right kidney, and pancreas were segmented manually on all 400 data sets, providing labeled ground truth data for training of a previously described U-Net-based deep learning framework for automated medical image segmentation (nnU-Net). The trained models were tested on all data sets using a 4-fold cross-validation scheme. Qualitative analysis of automated segmentation results was performed visually; performance metrics between automated and manual segmentation results were computed for quantitative analysis. In addition, interobserver segmentation variability between 2 human readers was assessed on a subset of the data. Automated abdominal organ segmentation was performed with high qualitative and quantitative accuracy on UKBB and GNC data. In more than 90% of data sets, no or only minor visually detectable qualitative segmentation errors occurred. Mean Dice scores of automated segmentations compared with manual reference segmentations were well higher than 0.9 for the liver, spleen, and kidneys on UKBB and GNC data and around 0.82 and 0.89 for the pancreas on UKBB and GNC data, respectively. Mean average symmetric surface distance was between 0.3 and 1.5 mm for the liver, spleen, and kidneys and between 2 and 2.2 mm for pancreas segmentation. The quantitative accuracy of automated segmentation was comparable with the agreement between 2 human readers for all organs on UKBB and GNC data. Automated segmentation of abdominal organs is possible with high qualitative and quantitative accuracy on whole-body MR imaging data acquired as part of UKBB and GNC. The results obtained and deep learning models trained in this study can be used as a foundation for automated analysis of thousands of MR data sets of UKBB and GNC and thus contribute to tackling topical and original scientific questions.

CT image segmentation for inflamed and fibrotic lungs using a multi-resolution convolutional neural network

The purpose of this study was to develop a fully-automated segmentation algorithm, robust to various density enhancing lung abnormalities, to facilitate rapid quantitative analysis of computed tomography images. A polymorphic training approach is proposed, in which both specifically labeled left and right lungs of humans with COPD, and nonspecifically labeled lungs of animals with acute lung injury, were incorporated into training a single neural network. The resulting network is intended for predicting left and right lung regions in humans with or without diffuse opacification and consolidation. Performance of the proposed lung segmentation algorithm was extensively evaluated on CT scans of subjects with COPD, confirmed COVID-19, lung cancer, and IPF, despite no labeled training data of the latter three diseases. Lobar segmentations were obtained using the left and right lung segmentation as input to the LobeNet algorithm. Regional lobar analysis was performed using hierarchical clustering to identify radiographic subtypes of COVID-19. The proposed lung segmentation algorithm was quantitatively evaluated using semi-automated and manually-corrected segmentations in 87 COVID-19 CT images, achieving an average symmetric surface distance of 0.495pm 0.309 mm and Dice coefficient of 0.985pm 0.011. Hierarchical clustering identified four radiographical phenotypes of COVID-19 based on lobar fractions of consolidated and poorly aerated tissue. Lower left and lower right lobes were consistently more afflicted with poor aeration and consolidation. However, the most severe cases demonstrated involvement of all lobes. The polymorphic training approach was able to accurately segment COVID-19 cases with diffuse consolidation without requiring COVID-19 cases for training.

Design of two-mirror telescope systems with freeform surfaces: modified configurations and analysis

Conventional two-mirror optical telescope designs are well known. An attempt to improve the performance of a two-mirror telescopic system using freeform surface is reported. Four variants of the optical design that use symmetric and off-axis freeform surfaces for achieving superior performances in the spectral range from 400 to 900 nm are proposed. These designs are compared with the conventional Ritchey–Chretien and equivalent two-mirror off-axis telescope designs with rotationally symmetric surfaces. The optical design with freeform surfaces shows marked improvements compared with its counterpart comprising of conics and higher order aspherics. The incorporation of freeform surfaces is obtained by an overlay of fringe Zernike polynomial either on the base sphere or on the conic itself, which is used as a surface descriptor in the envisaged designs. This approach aids in correction of asymmetrical aberrations and also extends the performances to a wider field, which is quite advantageous in the case of off-axis (de-centered and tilted) optical systems.

Development of a green synthesized silver nanoparticle-based antioxidant capacity method using carob extract

The measurement of total antioxidant capacity (TAC) indicating the cooperative action of all antioxidants in a complex sample is an important analytical challenge. A novel green silver nanoparticle-based antioxidant capacity (GSNP-AC) method using carob extract was designed for TAC measurement. Green synthesis of nanoparticles has various advantages such as environmentally friendliness, reliability, sustainability and fast production. In the presence of silver seeds formed with carob extract, Ag+ is reduced to spherical silver nanoparticles (SNPs) with antioxidants, resulting in enlarged particles with a symmetric surface plasmon resonance (SPR) absorption band at 434 nm. The SPR of SNPs allowed quantification of antioxidants (i.e., increase in SPR absorbance being correlated to antioxidant concentration), and the GSNP-AC method gave a linear response over a wide concentration range of antioxidants. The molar absorptivity and LOD for trolox were 11700 L mol−1 cm−1 and 0.31 µM, respectively. The proposed method was validated in terms of concentration-dependent linear response, additivity of absorbances, precision and accuracy. The results obtained with the proposed method for standard antioxidants, synthetic mixtures and real sample extracts were correlated with those of the traditional CUPRAC method. This green synthesized silver nanoparticle-based TAC method showed simple, rapid, cost-effective, and eco-friendly superiorities over similar other nanoparticle-based assays.

High sensitivity refractive index sensor in long-range surface plasmon resonance based on side polished optical fiber

A side-polished single mode optical fiber based on conventional long-range surface plasmon resonance (cLRSPR) sensor and symmetrical long-range surface plasmon resonance (sLRSPR) sensor was proposed and experimentally demonstrated. The proposed structure of cLRSPR is composed of optical fiber/dielectric buffer layer (DBL)/Ag film while for sLRSPR is consisted of optical fiber/DBL/Ag film/DBL. Magnesium fluoride, MgF2 with 50 nm thickness is used as DBL layer that lies beneath of Ag film in cLRSPR and covering both sides of Ag film in sLRSPR. 40 nm thickness of Ag layer is used with 5 mm length of sensing region. The result suggested that the sensing performance in terms of sensitivity of the cLRSPR sensor was better compared to sLRSPR and previous works of conventional surface plasmon resonance (cSPR) as it achieves as high as 2812.5 nm/RIU at 0.01 interval of refractive index with refractive index (RI) range of 1.33–1.34. The larger the sensitivity value indicates the better the performance of the SPR sensor. This design is expected to play an important role on chemical sensing and biological sensing as it displays higher penetration depth.

Combining Images and T-Staging Information to Improve the Automatic Segmentation of Nasopharyngeal Carcinoma Tumors in MR Images

The accurate and reproducible delineation of tumors from uninvolved tissue is essential for radiation oncology. However, the tumor margin may be challenging to identify from magnetic resonance (MR) images of nasopharyngeal carcinomas (NPCs). Additionally, clinical diagnoses such as T-staging may already provide some information on tumor invasion. To use this information and improve the performance of tumor segmentation, we propose a novel deep learning neural network architecture that can incorporate both T-staging and image information. Based on U-Net, our model adds a T-channel composed of T-staging information and uses the attention mechanism. Since the T-staging information is defined by the extent of tumor invasion, the T-channel using T-staging information can improve the segmentation accuracy at different stages. Additionally, the addition of an attention mechanism allows our model to retain the most valuable pixels of the image, thus further improving the delineation accuracy. In our experiments, the proposed network was trained and validated based on records from 251 clinical patients using 10-fold cross-validation. The dice similarity coefficient (DSC) and average symmetric surface distance (ASSD) were used to evaluate our network's results. The average DSC and ASSD and their standard deviation (SD) values are 0.841 ± 0.011 and 0.747 ± 0.199 mm. The unique T-channel effectively utilizes T-staging information to improve the results. With the combination of the T-channel module and the attention module, we significantly improved NPC tumor delineation performance.

Semantic Constraint Based Unsupervised Domain Adaptation for Cardiac Segmentation

The segmentation of unlabeled medical images is troublesome due to the high cost of annotation, and unsupervised domain adaptation is one solution to this. In this paper, an improved unsupervised domain adaptation method was proposed. The proposed method considered both global alignment and category-wise alignment. First, we aligned the appearance of two domains by image transformation. Second, we aligned the output maps of two domains in a global way. Then, we decomposed the semantic prediction map by category, aligning the prediction maps in a category-wise manner. Finally, we evaluated the proposed method on the 2017 Multi-Modality Whole Heart Segmentation Challenge dataset, and obtained 82.1 on the dice similarity coefficient and 4.6 on the average symmetric surface distance, demonstrating the effectiveness of the combination of global alignment and category-wise alignment.