Automatic Segmentation Algorithm Research Articles

A Deep Learning Algorithm for Automatic 3D Segmentation of Rotator Cuff Muscle and Fat from Clinical MRI Scans.

The authors aimed to develop and validate an automated artificial intelligence (AI) algorithm for three-dimensional (3D) segmentation of all four rotator cuff (RC) muscles to quantify intramuscular fat infiltration (FI) and individual muscle volume. The dataset included retrospectively collected RC MRI scans in 232 patients (63 with normal RCs, 169 with RC tears). A two-stage AI model was developed to segment all RC muscles and their FI in each stage. For comparison, single-stage and Otsu filtering models were created. Using the two-stage model, segmentation performance demonstrated high Dice scores (mean, 0.92 ± 0.14 [SD]), low volume errors (mean, 5.72% ± 9.23), and low FI errors (mean, 1.54% ± 2.79) when validated in 30 scans. There was a significant correlation between the 3D FI in the RC tear scans with a Goutallier grade (ρ = 0.53, P < .001) and FI found from a single two-dimensional (2D) section (all muscles, ρ > 0.70; P < .001). However, Bland-Altman analysis of the 3D compared with the 2D analyses of FI demonstrated a proportional bias (all muscles, P < .001). Compared with Goutallier classification or single-image quantification, the AI method allowed for more variability in images and led to objective separate quantifications of muscle volume and FI in all RC muscles. Keywords: Rotator Cuff, Artificial Intelligence, Segmentation, Fat Infiltration, Muscle Volume, MRI, Shoulder Supplemental material is available for this article. © RSNA, 2023.

Semi-Supervised Medical Image Segmentation Guided by Bi-Directional Constrained Dual-Task Consistency

Medical image processing tasks represented by multi-object segmentation are of great significance for surgical planning, robot-assisted surgery, and surgical safety. However, the exceptionally low contrast among tissues and limited available annotated data makes developing an automatic segmentation algorithm for pelvic CT challenging. A bi-direction constrained dual-task consistency model named PICT is proposed to improve segmentation quality by leveraging free unlabeled data. First, to learn more unmarked data features, it encourages the model prediction of the interpolated image to be consistent with the interpolation of the model prediction at the pixel, model, and data levels. Moreover, to constrain the error prediction of interpolation interference, PICT designs an auxiliary pseudo-supervision task that focuses on the underlying information of non-interpolation data. Finally, an effective loss algorithm for both consistency tasks is designed to ensure the complementary manner and produce more reliable predictions. Quantitative experiments show that the proposed PICT achieves 87.18%, 96.42%, and 79.41% mean DSC score on ACDC, CTPelvic1k, and the individual Multi-tissue Pelvis dataset with gains of around 0.8%, 0.5%, and 1% compared to the state-of-the-art semi-supervised method. Compared to the baseline supervised method, the PICT brings over 3-9% improvements. The developed PICT model can effectively leverage unlabeled data to improve segmentation quality of low contrast medical images. The segmentation result could improve the precision of surgical path planning and provide input for robot-assisted surgery.

Combining PropSeg and a convolutional neural network for automatic spinal cord segmentation in pediatric populations and patients with spinal cord injury

AbstractSegmentation of the spinal cord is an essential process for the accurate delineation of spinal cord structures but can be a tedious task for experts when using manual or semi‐automated tools. On the other hand, existing automatic segmentation algorithms have not been developed with the pediatric or injured spinal cord in mind. This study presents a novel automated segmentation method that combines the flexibility of deterministic approaches and the powerfulness of neural networks, applied to pediatric and injured spinal cord magnetic resonance imaging (MRI) data. The method first applies the PropSeg algorithm several times on small patches of the spinal cord MRI with various initialization parameters. Then, a convolutional neural network concatenates all these small segmentations with the original MR images to compute a final segmentation. Our results demonstrate good performances on the whole spinal cord (Dice score = 0.88 vs. 0.9) while outperforming existing methods on spinal cord injury regions (0.8 vs. 0.63).

Abstract 106: Novel Imaging-based Morphological Markers For Improved Prediction Of Stroke Outcomes: A Machine Learning Approach

Patient selection for acute reperfusion therapies of acute ischemic stroke is based on assessment of the core and tissue-at-risk and the time from stroke onset. Response to treatment and clinical outcomes, however, vary significantly since patient-specific cerebrovascular anatomy plays a vital role in governing flow and reperfusion. We developed an end-to-end automated machine learning approach to extract advanced cerebrovascular morphological features, including tortuosity and collateral index, to improve outcome prediction, and patient selection. Methods: Using our validated automatic cerebrovascular segmentation and feature extraction algorithm, we obtained the morphological properties of the brain vasculature, including vessel tortuosity, length, diameter, and bifurcation patterns, in MR angiography scans of 100 anonymized stroke patients. Each patient’s collateral index was also automatically graded using a previously established probabilistic atlas of healthy cerebrovasculature. To predict 90-day functional outcomes (mRS), a statistical model was trained using patients' clinical features (demographics, comorbidities, and baseline NIHSS), imaging features (initial ASPECTS, perfusion mismatch volume, and collateral index), and the extracted morphologic features (length, diameter, number of branches, vessel tortuosity, fractal dimension, and total volume). The predictions were compared against ground truth mRS to assess model accuracy. We also compared the model’s performance against conventional outcome predictors with only the clinical and imaging features. Results: The conventional model including the common clinical and imaging features excluding the collateral index gave an area under the curve (AUC) of 0.63, which is comparable to previously published results. Including the automatically scored collateral index improved the accuracy to 0.74. We found that further including the morphological features to the predictor model significantly improved AUC accuracy to 0.83. Conclusion: Including automatically extracted cerebrovascular morphologic features in traditional clinical and imaging markers significantly improves outcome prediction and response to treatment in stroke patients.

Abstract TP100: Transfer Learning Based Cerebral Microbleed Detection As An Mri Biomarker For Cerebral Amyloid Angiopathy Spectrum Diseases

Objectives: Quantifying cerebral microbleeds (CMBs) in cerebral amyloid angiopathy (CAA) is important, however manual counting of CMBs on brain MRI is laborious. This work aims to develop a novel automatic deep learning segmentation algorithm for fast and accurate quantification of CMBs on brain MRIs. Method: We used (a) internal dataset (susceptibility-weighted imaging [SWI] clinical scans from 83 patients with CAA) with heterogeneous protocols on 1.5T and 3T scanners and manual CMB segmentation by 2 readers, and (b) public dataset (57 SWI scans from 30 patients with Alzheimer’s and CMB locations). The data was split into training, validation, and testing subsets (internal: 50/20/13; Public: 38/9/10). Preprocessing included N3 bias field correction, brain extraction, and up-sampling to a fixed in-plane resolution (0.2mm). Model used multi-channel 2D input (Fig1.b) and mimicked manual CMB segmentation. We trained base-model initially using the public dataset and then applied transfer learning using the internal dataset to segment CMBs in clinical scans. We evaluated model’s performance by sensitivity (S), precision (P), number of false positives per scan (FPC) and F1-score (F1). Results: On 10 public test scans, base model performed well (S 0.81, P 0.74, FPC 0.60, F1 0.77) but had high FPC for 13 internal test scans (S 1.00, P 0.15, FPC 4.31, F1 0.26). After transfer learning, the precision, FPC and F1-score were improved for internal scans (S 0.80, P 0.80, FPC 0.15, F1 0.80). Conclusion: Refined deep-learning based CMB counting algorithms could quantify CMBs for heterogeneous clinical MRI dataset with good sensitivity, precision, and false positivity. This method may accelerate further CAA clinical diagnosis and research.

HMNet: Hierarchical Multi-Scale Brain Tumor Segmentation Network.

An accurate and efficient automatic brain tumor segmentation algorithm is important for clinical practice. In recent years, there has been much interest in automatic segmentation algorithms that use convolutional neural networks. In this paper, we propose a novel hierarchical multi-scale segmentation network (HMNet), which contains a high-resolution branch and parallel multi-resolution branches. The high-resolution branch can keep track of the brain tumor's spatial details, and the multi-resolution feature exchange and fusion allow the network's receptive fields to adapt to brain tumors of different shapes and sizes. In particular, to overcome the large computational overhead caused by expensive 3D convolution, we propose a lightweight conditional channel weighting block to reduce GPU memory and improve the efficiency of HMNet. We also propose a lightweight multi-resolution feature fusion (LMRF) module to further reduce model complexity and reduce the redundancy of the feature maps. We run tests on the BraTS 2020 dataset to determine how well the proposed network would work. The dice similarity coefficients of HMNet for ET, WT, and TC are 0.781, 0.901, and 0.823, respectively. Many comparative experiments on the BraTS 2020 dataset and other two datasets show that our proposed HMNet has achieved satisfactory performance compared with the SOTA approaches.

An effective approach for the nuclei segmentation from breast histopathological images using star-convex polygon

Early detection and patient-driven precise treatment can drastically reduce the elevation in the mortality rate due to breast tumours in women. Therefore, it is inevitable to develop an interpretable model, which detects and segments cancer cell nuclei. The main challenge to achieve this goal is the separation of the individual cell nucleus from the clustered overlapping nuclei without losing the pixel-level shape correspondence of each nucleus. The performance of the most common image processing conventional methods alone has not yielded the expected result for nuclei segmentation from the digitized tissue images either due to over-segmentation or under-segmentation. Here, this issue is addressed by developing an automatic nuclei segmentation algorithm which uses a deep learning-based approach. This paper introduces pixel-level analysis using the star convex polygon approach for segmenting the cell nuclei from the histopathological images. Analysis shows that this technique exemplifies a more accurate segmentation result with a validation IOU score of 87.56%. The true positive rate and the detection of pixel-level shape correspondence are also comparatively high for this method during the test phase.

Hypothalamic volume is associated with body mass index

The hypothalamus is an important neuroendocrine hub for the control of appetite and satiety. In animal studies it has been established that hypothalamic lesioning or stimulation causes alteration to feeding behaviour and consequently body mass, and exposure to high calorie diets induces hypothalamic inflammation. These findings suggest that alterations in hypothalamic structure and function are both a cause and a consequence of changes to food intake. However, there is limited in vivo human data relating the hypothalamus to obesity or eating disorders, in part due to technical problems relating to its small size. Here, we used a novel automated segmentation algorithm to exploratorily investigate the relationship between hypothalamic volume, normalised to intracranial volume, and body mass index (BMI). The analysis was applied across four independent datasets comprising of young adults (total n = 1,351 participants) spanning a range of BMIs (13.3 – 47.8 kg/m2). We compared underweight (including individuals with anorexia nervosa), healthy weight, overweight and obese individuals in a series of complementary analyses.We report that overall hypothalamic volume is significantly larger in overweight and obese groups of young adults. This was also observed for a number of hypothalamic sub-regions. In the largest dataset (the HCP-Young Adult dataset (n = 1111)) there was a significant relationship between hypothalamic volume and BMI.We suggest that our findings of a positive relationship between hypothalamic volume and BMI is potentially consistent with hypothalamic inflammation as seen in animal models in response to high fat diet, although more research is needed to establish a causal relationship. Overall, we present novel, in vivo findings that link elevated BMI to altered hypothalamic structure. This has important implications for study of the neural mechanisms of obesity in humans.

Automatic Image Guidance for Assessment of Placenta Location in Ultrasound Video Sweeps.

Ultrasound-based assistive tools are aimed at reducing the high skill needed to interpret a scan by providing automatic image guidance. This may encourage uptake of ultrasound (US) clinical assessments in rural settings in low- and middle-income countries (LMICs), where well-trained sonographers can be scarce. This paper describes a new method that automatically generates an assistive video overlay to provide image guidance to a user to assess placenta location. The user captures US video by following a sweep protocol that scans a U-shape on the lower maternal abdomen. The sweep trajectory is simple and easy to learn. We initially explore a 2-D embedding of placenta shapes, mapping manually segmented placentas in US video frames to a 2-D space. We map 2013 frames from 11 videos. This provides insight into the spectrum of placenta shapes that appear when using the sweep protocol. We propose classification of the placenta shapes from three observed clusters: complex, tip and rectangular. We use this insight to design an effective automatic segmentation algorithm, combining a U-Net with a CRF-RNN module to enhance segmentation performance with respect to placenta shape. The U-Net+CRF-RNN algorithm automatically segments the placenta and maternal bladder. We assess segmentation performance using both area and shape metrics. We report results comparable to the state-of-the-art for automatic placenta segmentation on the Dice metric, achieving 0.83 ± 0.15 evaluated on 2127 frames from 10 videos. We also qualitatively evaluate 78,308 frames from 135 videos, assessing if the anatomical outline is correctly segmented. We found that addition of the CRF-RNN improves over a baseline U-Net when faced with a complex placenta shape, which we observe in our 2-D embedding, up to 14% with respect to the percentage shape error. From the segmentations, an assistive video overlay is automatically constructed that (i) highlights the placenta and bladder, (ii) determines the lower placenta edge and highlights this location as a point and (iii) labels a 2-cm clearance on the lower placenta edge. The 2-cm clearance is chosen to satisfy current clinical guidelines. We propose to assess the placenta location by comparing the 2-cm region and the bottom of the bladder, which represents a coarse localization of the cervix. Anatomically, the bladder must sit above the cervix region. We present proof-of-concept results for the video overlay.

Automatic segmentation algorithm for high-spatial-resolution remote sensing images based on self-learning super-pixel convolutional network

ABSTRACT Super-pixel algorithms based on convolutional neural networks with fuzzy C-means clustering are widely used for high-spatial-resolution remote sensing images segmentation. However, this model requires the number of clusters to be set manually, resulting in a low automation degree due to the complexity of the iterative clustering process. To address this problem, a segmentation method based on a self-learning super-pixel network (SLSP-Net) and modified automatic fuzzy clustering (MAFC) is proposed. SLSP-Net performs feature extraction, non-iterative clustering, and gradient reconstruction. A lightweight feature embedder is adopted for feature extraction, thus expanding the receiving range and generating multi-scale features. Automatic matching is used for non-iterative clustering, and the overfitting of the network model is overcome by adaptively adjusting the gradient weight parameters, providing a better irregular super-pixel neighborhood structure. An optimized density peak algorithm is adopted for MAFC. Based on the obtained super-pixel image, this maximizes the robust decision-making interval, which enhances the automation of regional clustering. Finally, prior entropy fuzzy C-means clustering is applied to optimize the robust decision-making and obtain the final segmentation result. Experimental results show that the proposed model offers reduced experimental complexity and achieves good performance, realizing not only automatic image segmentation, but also good segmentation results.

Volume and complexity of the thalamus in Anorexia Nervosa: An exploratory evaluation.

Recent neuroscientific findings have highlighted the role of the thalamus in several cognitive functions, ranging from perception to cognitive flexibility, memory, and body representation. Since some of these functions may be involved in the pathophysiology of Anorexia Nervosa (AN), this study aims at exploring thalamic structure in different phases of the disorder. The sample included 38 patients with acute AN, 20 patients who fully recovered from AN (recAN), and 38 healthy controls (HC), all female. All participants underwent high-resolution MRI. The volumes of the whole thalamus and 25 thalamic nuclei were extracted using an automated segmentation algorithm, and thalamic fractal dimension was estimated using the calcFD toolbox. Patients with acute AN, compared to HC, displayed reduced thalamic volume and complexity both at the whole level and at the level of specific nuclei. In patients recAN, instead, alterations were observed only at the level of the right laterodorsal and central lateral nuclei. In the acute phase of the disorder patients with AN present a widespread reduction in thalamic volume and complexity. However, these alterations seem to normalise almost completely following weight restoration, thus suggesting the involvement of malnutrition-related mechanisms.

Thyroid region prior guided attention for ultrasound segmentation of thyroid nodules

Ultrasound segmentation of thyroid nodules is a challenging task, which plays an vital role in the diagnosis of thyroid cancer. However, the following two factors limit the development of automatic thyroid nodule segmentation algorithms: (1) existing automatic nodule segmentation algorithms that directly apply semantic segmentation techniques can easily mistake non-thyroid areas as nodules, because of the lack of the thyroid gland region perception, the large number of similar areas in the ultrasonic images, and the inherently low contrast images; (2) the currently available dataset (i.e., DDTI) is small and collected from a single center, which violates the fact that thyroid ultrasound images are acquired from various devices in real-world situations. To overcome the lack of thyroid gland region prior knowledge, we design a thyroid region prior guided feature enhancement network (TRFE+) for accurate thyroid nodule segmentation. Specifically, (1) a novel multi-task learning framework that simultaneously learns the nodule size, gland position, and the nodule position is designed; (2) an adaptive gland region feature enhancement module is proposed to make full use of the thyroid gland prior knowledge; (3) a normalization approach with respect to the channel dimension is applied to alleviate the domain gap during the training process. To facilitate the development of thyroid nodule segmentation, we have contributed TN3K: an open-access dataset containing 3493 thyroid nodule images with high-quality nodule masks labeling from various devices and views. We perform a thorough evaluation based on the TN3K test set and DDTI to demonstrate the effectiveness of the proposed method. Code and data are available at https://github.com/haifangong/TRFE-Net-for-thyroid-nodule-segmentation.

Automatic 3D CT liver segmentation based on fast global minimization of probabilistic active contour.

Automatic liver segmentation from computed tomography (CT) images is an essential preprocessing step for computer-aided diagnosis of liver diseases. However, due to the large differences in liver shapes, low-contrast to adjacent tissues, and existence of tumors or other abnormalities, liver segmentation has been very challenging. This study presents an accurate and fast liver segmentation method based on a novel probabilistic active contour (PAC) model and its fast global minimization scheme (3D-FGMPAC), which is explainable as compared with deep learningmethods. The proposed method first constructs a slice-indexed-histogram to localize the volume of interest (VOI) and estimate the probability that a voxel belongs to the liver according its intensity. The probabilistic image would be used to initialize the 3D PAC model. Secondly, a new contour indicator function, which is a component of the model, is produced by combining the gradient-based edge detection and Hessian-matrix-based surface detection. Then, a fast numerical scheme derived for the 3D PAC model is performed to evolve the initial probabilistic image into the global minimizer of the model, which is a smoothed probabilistic image showing a distinctly highlighted liver. Next, a simple region-growing strategy is applied to extract the whole liver mask from the smoothed probabilistic image. Finally, a B-spline surface is constructed to fit the patch of the rib cage to prevent possible leakage into adjacent intercostaltissues. The proposed method is evaluated on two public datasets. The average Dice score, volume overlap error, volume difference, symmetric surface distance and volume processing time are 0.96, 7.35%, 0.02%, 1.17 mm and 19.8 s for the Sliver07 dataset, and 0.95, 8.89%, , 1.45 mm and 23.08 s for the 3Dircadb dataset,respectively. The proposed fully-automatic approach can effectively segment the liver from low-contrast and complex backgrounds. The quantitative and qualitative results demonstrate that the proposed segmentation method outperforms state-of-the-art traditional automatic liver segmentation algorithms and achieves very competitive performance compared with recent deep leaning-basedmethods.

Quantitative Evaluation of Changes in Three-Dimensional CT Density Distributions in Pulmonary Alveolar Proteinosis after GM-CSF Inhalation

Background: A previous clinical trial for autoimmune pulmonary alveolar proteinosis (APAP) demonstrated that granulocyte-macrophage colony-stimulating factor (GM-CSF) inhalation reduced the mean density of the lung field on computed tomography (CT) across 18 axial slice planes at a two-dimensional level. In contrast, in this study, we challenged three-dimensional analysis for changes in CT density distribution using the same datasets. Methods: As a sub-study of the trial, CT data of 31 and 27 patients who received GM-CSF and placebo, respectively, were analyzed. To overcome the difference between various shooting conditions, a newly developed automatic lung field segmentation algorithm was applied to CT data to extract the whole lung volume, and the accuracy of the segmentation was evaluated by five pulmonary physicians independently. For normalization, the percent pixel (PP) in a certain density range was calculated as a percentage of the total number of pixels from −1,000 to 0 HU. Results: The automatically segmented images revealed that the lung field was accurately extracted except for 7 patients with minor deletion or addition. Using the change in PP from baseline to week 25 (ΔPP) as the vertical axis, we created a histogram with 143 HU bins set for each patient. The most significant difference in ΔPP between GM-CSF and placebo groups was observed in two ranges: from −1,000 to −857 and −143 to 0 HU. Conclusion: Whole lung extraction followed by density histogram analysis of ΔPP may be an appropriate evaluation method for assessing CT improvement in APAP.

Automated Segmentation and Morphological Characterization of Hepatic Steatosis and Correlation with Histopathology

Prevalence of nonalcoholic fatty liver disease (NAFLD) has increased to 25% of the world population. Hepatic steatosis is a hallmark feature of NAFLD and is assessed histologically usingvisual and ordinal fat grading criteria (0-3) from the Nonalcoholic Steatohepatitis (NASH) Clinical Research Network (CRN) scoring system. The purpose of this study is to automatically segment and extract morphological characteristics and distributions of fat droplets (FDs) on liver histology images and find associations with severity of steatosis. A previously published human cohort of 68 NASH candidates was graded for steatosis by an experienced pathologist using the Fat CRN grading system. The automated segmentation algorithm quantified fat fraction (FF) and fat-affected hepatocyte ratio (FHR), extracted fat morphology by calculating radius and circularity of FDs, and examined FDs distribution and heterogeneity using nearest neighbor distance and regional isotropy. Regression analysis and Spearman correlation (ρ) yielded high correlations for radius (R2=0.86, ρ=0.72), nearest neighbor distance (R2=0.82, ρ=-0.82), regional isotropy (R2=0.84, ρ=0.74), and FHR (R2=0.90, ρ=0.85), and low correlation for circularity (R2=0.48, ρ=-0.32) with FF and pathologist grades, respectively. FHR showed a better distinction between pathologist Fat CRN grades compared to conventional FF measurements, making it a potential surrogate measure for Fat CRN scores. Our results showed variation in distribution of morphological features and steatosis heterogeneity within the same patient's biopsy sample as well as between patients of similar FF. The fat percentage measurements, specific morphological characteristics, and patterns of distribution quantified with the automated segmentation algorithm showed associations with steatosis severity; however, future studies are warranted to evaluate the clinical significance of these steatosis features in progression of NAFLD and NASH.

Automatic Segmentation Algorithm of Dermoscopy Image Based on Transformer and Convolutional Neural Network

The lesion area of the dermoscopic image is highly similar to the background pixels, and there are various shapes, blurred edges, artificial or hair occlusion, etc. In order to obtain higher-precision segmentation of skin lesions, a automatic segmentation algorithm for dermoscopic images is proposed in this paper. Firstly, ResNet 34 is used to extract multiple resolution features, and the Transformer module is used to model globally the input features in the context part. Secondly, the multi-scale information of context features is aggregated through the Mixed Pooling Module. In addition, an efficient convolution module is designed between jump links of the corresponding codec modules to improve the edge refinement and anti-interference ability of the jump path. Finally, the decoder is used to restore the image resolution and fuse other shallow resolution features information, and using Focal Loss function to improve the accuracy of segment targets. The Dice coefficient, accuracy, Jackard index and sensitivity scores obtained by our method on the ISIC 2017 and ISIC 2018 datasets are 88.83%, 94.77%, 81.43%, 88.49% and 89.46%, 94.50%, 82.56% and 94.62% respectively. Compared with other algorithms, our method has certain advantages, proving its effectiveness.

How white matter hyperintensities impacts Alzheimer's disease and frontotemporal dementia across underrepresented Latin American samples

AbstractBackgroundWhite matter hyperintensities (WMH) are frequently observed on MRI scans of older people. Usually interpreted as a sign of cerebrovascular disease (CVD), they are also associated with increased risk of cognitive impairment and dementia. While WMH and CVD are highly prevalent in Alzheimer's disease (AD), this has been less studied in behavioral variant frontotemporal dementia (bvFTD), letting alone their study in underrepresented populations with increased CVD. Here, we investigated WMH in AD and bvFTD patients in Latin America (Argentina, Chile and Colombia), a region with high prevalence of CVD. We analyzed total volume and regional distribution of WMH and its association with cognitive impairment and brain lacunes, a marker of CVD.MethodA total of 66 patients (43 AD, 23 bvFTD) and 66 controls (CN) from the ReDLat consortium underwent high‐resolution brain MRI and neuropsychological examination. WMH were extracted from Flair images using automated segmentation algorithms to perform analyses of total volume and spatial distribution. Brain lacunes were visually quantified from T1 and Flair images. Group differences in total and regional WMH volume and associations with different cognitive domains, age, and brain lacunes were examined.ResultTotal and regional WMH volumes were significant larger in AD patients compared to CN and to bvFTD. No differences were found between CN and bvFTD. Only in AD, regional WMH load correlated with worse performance in a particular executive functioning task. No differences were found in the proportion of lacunes among groups nor in the total load of WMH of subjects with or without lacunes. In CN and AD, but not FTD, there was a positive correlation between total WMH volume and age, with a differential topographic distribution with a selective involvement of the bilateral cingulum‐hippocampus bundle in AD.ConclusionEnhanced WMH was observed in AD but not in FTD. WMH loading in specific brain regions contribute to domain‐specific cognitive deficits in AD. WMH could be partially independent of vascular pathology (it was unrelated to brain lacunes) suggesting a specific neurodegenerative origin in AD. The typical earlier onset of brain abnormalities in FTD could mask the age‐related increase in WMH.

Spleen Tissue Segmentation Algorithm for Cryo-Imaging Data.

Spleen tissue segmentation is an essential process for analyzing various immunological diseases as observed in the cryo-imaging data. Because manual labeling of the spleen tissue by human experts is not efficient, an automatic segmentation algorithm is needed. In this study, we developed a novel algorithm for automatically segmenting spleen substructures including white pulp and red pulp for the first time. The algorithm is designed for datasets created by a cryo-imaging system. This unique technology can effectively enable cellular tracking anywhere in the whole mouse with single-cell sensitivity. The proposed algorithm consists of four components: initial spleen mask creation, feature extraction, Supervised Patch-based Fuzzy c-Mean (spFCM) classification, and post-processing. The algorithm accurately and efficiently labeled spleen tissues in all experiment settings. The algorithm also improved the spleen segmentation throughput by 90 folds as compared to the manual segmentation. Moreover, we show that our novel spFCM algorithm outperformed traditional fast-learning classifiers as well as the U-Net deep-learning model in many aspects. Two major contributions of this paper are (1) an explainable algorithm for segmenting spleen tissues in cryo-images for the first time and (2) an spFCM algorithm as a new classifier. We also discussed that our work can be beneficial to researchers who work not only in the fields of graft-versus-host disease (GVHD) mouse models, but also in that of other immunological disease models where spleen analysis is essential. Future work building upon our research may lay the foundations for biomedical studies that utilize cryo-imaging technology.

MTNet: A combined diagnosis algorithm of vessel segmentation and diabetic retinopathy for retinal images.

Medical studies have shown that the condition of human retinal vessels may reveal the physiological structure of the relationship between age-related macular degeneration, glaucoma, atherosclerosis, cataracts, diabetic retinopathy, and other ophthalmic diseases and systemic diseases, and their abnormal changes often serve as a diagnostic basis for the severity of the condition. In this paper, we design and implement a deep learning-based algorithm for automatic segmentation of retinal vessel (CSP_UNet). It mainly adopts a U-shaped structure composed of an encoder and a decoder and utilizes a cross-stage local connectivity mechanism, attention mechanism, and multi-scale fusion, which can obtain better segmentation results with limited data set capacity. The experimental results show that compared with several existing classical algorithms, the proposed algorithm has the highest blood vessel intersection ratio on the dataset composed of four retinal fundus images, reaching 0.6674. Then, based on the CSP_UNet and introducing hard parameter sharing in multi-task learning, we innovatively propose a combined diagnosis algorithm vessel segmentation and diabetic retinopathy for retinal images (MTNet). The experiments show that the diagnostic accuracy of the MTNet algorithm is higher than that of the single task, with 0.4% higher vessel segmentation IoU and 5.2% higher diagnostic accuracy of diabetic retinopathy classification.

NIMG-72. VALIDATION OF CONDITIONAL AND SUPER-RESOLUTION GENERATIVE ADVERSARIAL NETWORKS FOR IMPUTATION OF MISSING BRAIN MRI SEQUENCES

Abstract Missing MRI sequences are common in radiology due to differences in protocols, time constraints, and image artifacts. Missing sequences are a nuisance for subsequent algorithms, such as automated segmentation algorithms requiring multiple MRI sequences (e.g., HD-GLIO, which requires T2, T1, post-contrast T1 (T1c), and FLAIR). Accurate tumor segmentation is a fundamental step in extracting radiomics analysis to use, for example, to predict the IDH status of glioma or their overall survival. We validated a conditional (pix2pix) and super-resolution GAN (SOUP-GAN) for imputing MRI sequences on 285 adult glioma with complete MRI sequences available. We simulated missing T1 sequences and subsequently imputed them from T1c using pix2pix and from sagittal T1 using SOUP-GAN. Using HD-GLIO, we obtained the true segmentations from the complete MRI sequences. HD-GLIO was also used after replacing the original T1 sequence with the one imputed with pix2pix and SOUP-GAN. Thus, there were three sets of segmentations for each subject. We compared T2 and T1c radiomics features extracted with Pyradiomics using segmentations obtained with complete data with features extracted from segmentation obtained with the imputed T1 sequences using intraclass correlation coefficient (ICC) and Bland-Altman (BA) analysis. Median (interquartile range) ICC for shape features extracted from T2 and T1c were 0.82(0.76-0.92) using pix2pix and 0.89(0.86-0.94) using SOUP-GAN. Median ICC for first-order-statistics features were 0.90(0.83-0.93) for T2 and 0.90(0.88-0.92) for T1c using pix2pix, and 0.61(0.22-0.82) for T2 and 0.42(0.34-0.57) for T1c using SOUP-GAN. Median ICC for texture features were 0.86(0.82-0.87) for T2 and 0.87(0.79-0.90) for T1c using pix2pix, and 0.69(0.47-0.81) for T2 and 0.76(0.65-0.83) for T1c using SOUP-GAN. We show that T2 and T1c shape features (e.g., volumetric analysis) can be adequately extracted from segmentations obtained with an algorithm requiring multiple MRI sequences using T1 sequences imputed with GANs; however, the effect of imputation differed for first-order-statistics and texture features.