Deep Learning Segmentation Methods Research Articles

Two fully automated data-driven 3D whole-breast segmentation strategies in MRI for MR-based breast density using image registration and U-Net with a focus on reproducibility

Presence of higher breast density (BD) and persistence over time are risk factors for breast cancer. A quantitatively accurate and highly reproducible BD measure that relies on precise and reproducible whole-breast segmentation is desirable. In this study, we aimed to develop a highly reproducible and accurate whole-breast segmentation algorithm for the generation of reproducible BD measures. Three datasets of volunteers from two clinical trials were included. Breast MR images were acquired on 3 T Siemens Biograph mMR, Prisma, and Skyra using 3D Cartesian six-echo GRE sequences with a fat-water separation technique. Two whole-breast segmentation strategies, utilizing image registration and 3D U-Net, were developed. Manual segmentation was performed. A task-based analysis was performed: a previously developed MR-based BD measure, MagDensity, was calculated and assessed using automated and manual segmentation. The mean squared error (MSE) and intraclass correlation coefficient (ICC) between MagDensity were evaluated using the manual segmentation as a reference. The test-retest reproducibility of MagDensity derived from different breast segmentation methods was assessed using the difference between the test and retest measures (Δ2-1), MSE, and ICC. The results showed that MagDensity derived by the registration and deep learning segmentation methods exhibited high concordance with manual segmentation, with ICCs of 0.986 (95%CI: 0.974-0.993) and 0.983 (95%CI: 0.961-0.992), respectively. For test-retest analysis, MagDensity derived using the registration algorithm achieved the smallest MSE of 0.370 and highest ICC of 0.993 (95%CI: 0.982-0.997) when compared to other segmentation methods. In conclusion, the proposed registration and deep learning whole-breast segmentation methods are accurate and reliable for estimating BD. Both methods outperformed a previously developed algorithm and manual segmentation in the test-retest assessment, with the registration exhibiting superior performance for highly reproducible BD measurements.

Deep active learning for suggestive segmentation of biomedical image stacks via optimisation of Dice scores and traced boundary length.

Manual segmentation of stacks of 2D biomedical images (e.g., histology) is a time-consuming task which can be sped up with semi-automated techniques. In this article, we present a suggestive deep active learning framework that seeks to minimise the annotation effort required to achieve a certain level of accuracy when labelling such a stack. The framework suggests, at every iteration, a specific region of interest (ROI) in one of the images for manual delineation. Using a deep segmentation neural network and a mixed cross-entropy loss function, we propose a principled strategy to estimate class probabilities for the whole stack, conditioned on heterogeneous partial segmentations of the 2D images, as well as on weak supervision in the form of image indices that bound each ROI. Using the estimated probabilities, we propose a novel active learning criterion based on predictions for the estimated segmentation performance and delineation effort, measured with average Dice scores and total delineated boundary length, respectively, rather than common surrogates such as entropy. The query strategy suggests the ROI that is expected to maximise the ratio between performance and effort, while considering the adjacency of structures that may have already been labelled - which decrease the length of the boundary to trace. We provide quantitative results on synthetically deformed MRI scans and real histological data, showing that our framework can reduce labelling effort by up to 60-70% without compromising accuracy.

Bayesian Deep Learning for Segmentation for Autonomous Safe Planetary Landing

Hazard detection is critical for enabling autonomous landing on planetary surfaces. Current state-of-the-art methods leverage traditional computer vision approaches to automate the identification of safe terrain from input digital elevation models (DEMs). However, performance for these methods can degrade for input DEMs with increased sensor noise. In the last decade, deep learning techniques have been developed for various applications. Nevertheless, their applicability to safety-critical space missions has often been limited due to concerns regarding their outputs’ reliability. In response to these limitations, this paper proposes an application of the Bayesian deep learning segmentation method for hazard detection. The developed approach enables reliable, safe landing site detection by i) generating simultaneously a safety prediction map and its uncertainty map via Bayesian deep learning and semantic segmentation, and ii) using the uncertainty map to filter out the uncertain pixels in the prediction map so that the safe site identification is performed only based on the certain pixels (i.e., pixels for which the model is certain about its safety prediction). Experiments are presented with simulated data based on a Mars HiRISE digital terrain model by varying uncertainty threshold and noise levels to demonstrate the performance of the proposed approach.

ISA-Net: Improved spatial attention network for PET-CT tumor segmentation

Background and Objective: Achieving accurate and automated tumor segmentation plays an important role in both clinical practice and radiomics research. Segmentation in medicine is now often performed manually by experts, which is a laborious, expensive and error-prone task. Manual annotation relies heavily on the experience and knowledge of these experts. In addition, there is much intra- and interobserver variation. Therefore, it is of great significance to develop a method that can automatically segment tumor target regions. Methods: In this paper, we propose a deep learning segmentation method based on multimodal positron emission tomography-computed tomography (PET-CT), which combines the high sensitivity of PET and the precise anatomical information of CT. We design an improved spatial attention network(ISA-Net) to increase the accuracy of PET or CT in detecting tumors, which uses multi-scale convolution operation to extract feature information and can highlight the tumor region location information and suppress the non-tumor region location information. In addition, our network uses dual-channel inputs in the coding stage and fuses them in the decoding stage, which can take advantage of the differences and complementarities between PET and CT. Results: We validated the proposed ISA-Net method on two clinical datasets, a soft tissue sarcoma(STS) and a head and neck tumor(HECKTOR) dataset, and compared with other attention methods for tumor segmentation. The DSC score of 0.8378 on STS dataset and 0.8076 on HECKTOR dataset show that ISA-Net method achieves better segmentation performance and has better generalization. Conclusions: The method proposed in this paper is based on multi-modal medical image tumor segmentation, which can effectively utilize the difference and complementarity of different modes. The method can also be applied to other multi-modal data or single-modal data by proper adjustment.

Analysis of Convolutional Neural Network Segmentation Algorithm of Adenomyoma.

Adenomoma is a common disease occurring in the female uterus. The symptoms and pain of adenomoma seriously troubled the physical and mental health of contemporary women. However, because of the significant advantages of nondestructive and low price, ultrasound examination is used as the main imaging method for clinical diagnosis of gynecological diseases at the present stage and is often used in the initial screening and postoperative diagnosis and treatment of uterine diseases. Imaging provides a very rich information in the medical diagnosis of tumor and is a very important basis for the disease diagnosis and treatment at this stage. Ultrasound images are different from medical images such as X-ray and MRI. Because of the characteristics of imaging principles and noise interference, ultrasound images need to rely on rich clinical experience of doctors in the process of disease diseases, which increases the difficulty and work burden of doctors to some extent. Therefore, the project aims to study the deep learning segmentation method suitable for ultrasonic images. Combined with the Deeplab network in the convolutional neural network, comparing the results of the FCN network, and then finding that the Deeplab network has obvious advantages as an image segmentation model of uterine adenomyoma. In clinical practice, it can reduce the work burden of doctors and try in the direction of uterine adenomyomas ultrasound image segmentation, to fill the gap in this field.

Analysis on Methods for Detecting Brain Tumor from MRI images

Abstract: Brain tumor detection from MRI images is a challenging process due to high diversity in the tumor pixels of different peoples. Automatic detection has got wide spread acclaim because the manual detection by experts is time consuming and prone to error in judjement.Due to its high mortality rate, detection of tumor automatically is a new emerging technique in bio medical imaging. Here we present a review of few methods from simple threholding to advanced deep learning methods for segmentation of tumor from MRI data. The segmentation of tumor methods is classified to image segmentation using gray level processing, machine learning and deep learning. We compare the result of various method in an attempt to find the best methods available. As medical imaging methods have improving day by day we hope this review will help to understand emerging trends in brain tumor detection.

S2S-ARSNet: Sequence-to-Sequence automatic renal segmentation network

Accurate segmentation of kidney contours in diuretic renography has important implications for clinical diagnosis and treatments. However, the lack of clear boundaries and high-quality images makes automatic segmentation challenging. This paper proposes a novel automatic renal segmentation network, S2S-ARSNet, combining Convolutional Long Short-Term Memory (ConvLSTM) with the Unet structure. Unet is used to learn the spatial information of each sequence, and ConvLSTM is used to discover the temporal information between sequences and automatically update the temporal state of the sequence. Moreover, an additional pre-trained Unet is applied to generate coarse masks at different times to simulate the displacement that may occur during the detection process. In this way, the spatiotemporal information is modelled, and all the information of the 3D data is fully utilized to eliminate false positives and improve the segmentation accuracy. Extensive experiments were performed on the diuretic renography dataset. The experimental results show that the proposed method can significantly enhance the kidney segmentation performance compared with other single-image-based deep learning segmentation methods.

Applying convolutional neural networks for detecting wheat stripe rust transmission centers under complex field conditions using RGB-based high spatial resolution images from UAVs

The use of unmanned aerial vehicle (UAV) provide a timely and low-cost means of accessing high spatial resolution imagery for crop disease detection. In this study, convolutional neural networks (CNNs) and RGB-based high spatial resolution images from UAVs were explored to detect wheat stripe rust transmission centers (Infected area accounted less than 1.35 %) occurrence in complex fields conditions in Hubei, China. To take full advantage of end-to-end learning capabilities, CNNs semantic segmentation architecture (deeplabv3+) was applied to per-pixel classify the imagery for the detection of healthy wheat and stripe-rust-infected wheat (SRIW). Using a rich dataset with diverse field conditions and sunlight illumination properties, we were able to accurately detect SRIW (Rust class F1 = 0.81). The study also evaluated the impact of classification framework and spatial resolution on model training. It revealed that the model accuracies improved for the rust class when the multi-branching binary framework instead of the multi-classification framework for CNN training with unbalanced classes. A coarser spatial resolution (8 cm) significantly decreased the model accuracy (Rust class F1-score). In addition, the Macro-disease index(MDI) was defined to quantitatively measure the occurrence of SRIW. Our results demonstrate the capability of ultra-high spatial resolution UAV imaging in detecting SRIW. With the end-to-end deep learning segmentation method greatly reducing the need for intensive preprocessing, the combination of CNNs and RGB-based ultra-high spatial resolution images from UAVs provides a simple and rapid method for accurate detection of crop disease on a large scale.

Abstract 1719: Investigating intra-tumor heterogeneity using multiplexed immunohistochemistry & deep learning: A new approach to spatially map the tumor microenvironment

Abstract The tumor microenvironment (TME) is a highly complex mixture containing epithelium, stroma and a diverse network of immune cells & the spatial organization of these immune cells within the TME reflects a crucial process in anti-tumor immunity. At present, the usual standard of care for assessing if a patient has cancer, its stage and its likely future biological behaviour is visual examination of one or more stained sections. Although recent advances in multiple immunostaining have enabled characterization of several parameters on a single tissue section. For a higher dimensional chromogen based methodology, we have developed a multiplexed IHC (mIHC) procedure combining multiple labels per round with several sequential rounds, enabling analysis of complex immune cell populations on a single slide through consecutive cycles of staining, destaining & hyperspectral imaging. By integrating serial imaging, sequential labeling & image registration, we are able to spatially map the TME. The process presented is using absorption microscopy, enabling these images to be done in a reasonable time frame. Robust, accurate, segmentation of cell nuclei for overlapping nuclei is one of the most significant unsolved issues in digital pathology. Using the counterstain concentration images from our mIHC methodology, we have trained a deep learning segmentation method to accurately segment individual cell nuclei within overlapping clusters of nuclei. By combining a multiplexed IHC technique which enables the detection of multiple markers on a single slide with deep learning segmentation methods to segment every individual cell nuclei in tissue sections with an accuracy comparable to human annotation. We can analyze the cell-cell interactions between immune and tumour cells, enhancing our ability to perform molecularly based single cell analysis of multiple cell types simultaneously within the tissue. These two techniques joined can be scaled up to the entire tissue section level, improving our understanding of the biological aggressiveness of specific cancers & enabling an accurate spatial cell level representation of the tissue. Citation Format: Kouther Noureddine, Paul Gallagher, Martial Guillaud, Calum MacAulay. Investigating intra-tumor heterogeneity using multiplexed immunohistochemistry & deep learning: A new approach to spatially map the tumor microenvironment [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1719.

Analytical Approaches for the Segmentation of the Zebrafish Brain Vasculature.

With advancements in imaging techniques, data visualization allows new insights into fundamental biological processes of development and disease. However, although biomedical science is heavily reliant on imaging data, interpretation of datasets is still often based on subjective visual assessment rather than rigorous quantitation. This overview presents steps to validate image processing and segmentation using the zebrafish brain vasculature data acquired with light sheet fluorescence microscopy as a use case. Blood vessels are of particular interest to both medical and biomedical science. Specific image enhancement filters have been developed that enhance blood vessels in imaging data prior to segmentation. Using the Sato enhancement filter as an example, we discuss how filter application can be evaluated and optimized. Approaches from the medical field such as simulated, experimental, and augmented datasets can be used to gain the most out of the data at hand. Using such datasets, we provide an overview of how biologists and data analysts can assess the accuracy, sensitivity, and robustness of their segmentation approaches that allow extraction of objects from images. Importantly, even after optimization and testing of a segmentation workflow (e.g., from a particular reporter line to another or between immunostaining processes), its generalizability is often limited, and this can be tested using double-transgenic reporter lines. Lastly, due to the increasing importance of deep learning networks, a comparative approach can be adopted to study their applicability to biological datasets. In summary, we present a broad methodological overview ranging from image enhancement to segmentation with a mixed approach of experimental, simulated, and augmented datasets to assess and validate vascular segmentation using the zebrafish brain vasculature as an example. © 2022 The Authors. Current Protocols published by Wiley Periodicals LLC. HIGHLIGHTS: Simulated, experimental, and augmented datasets provide an alternative to overcome the lack of segmentation gold standards and phantom models for zebrafish cerebrovascular segmentation. Direct generalization of a segmentation approach to the data for which it was not optimized (e.g., different transgenics or antibody stainings) should be treated with caution. Comparison of different deep learning segmentation methods can be used to assess their applicability to data. Here, we show that the zebrafish cerebral vasculature can be segmented with U-Net-based architectures, which outperform SegNet architectures.

PO-1787 Impact of deep learning segmentation methods on the robustness of MR glioblastoma radiomics

PO-1787 Impact of deep learning segmentation methods on the robustness of MR glioblastoma radiomics

Deep learning segmentation and quantification method for assessing epicardial adipose tissue in CT calcium score scans

Epicardial adipose tissue volume (EAT) has been linked to coronary artery disease and the risk of major adverse cardiac events. As manual quantification of EAT is time-consuming, requires specialized training, and is prone to human error, we developed a deep learning method (DeepFat) for the automatic assessment of EAT on non-contrast low-dose CT calcium score images. Our DeepFat intuitively segmented the tissue enclosed by the pericardial sac on axial slices, using two preprocessing steps. First, we applied a HU-attention-window with a window/level 350/40-HU to draw attention to the sac and reduce numerical errors. Second, we applied a novel look ahead slab-of-slices with bisection (“bisect”) in which we split the heart into halves and sequenced the lower half from bottom-to-middle and the upper half from top-to-middle, thereby presenting an always increasing curvature of the sac to the network. EAT volume was obtained by thresholding voxels within the sac in the fat window (− 190/− 30-HU). Compared to manual segmentation, our algorithm gave excellent results with volume Dice = 88.52% ± 3.3, slice Dice = 87.70% ± 7.5, EAT error = 0.5% ± 8.1, and R = 98.52% (p < 0.001). HU-attention-window and bisect improved Dice volume scores by 0.49% and 3.2% absolute, respectively. Variability between analysts was comparable to variability with DeepFat. Results compared favorably to those of previous publications.

FastSurferVINN: Building resolution-independence into deep learning segmentation methods—A solution for HighRes brain MRI

Leading neuroimaging studies have pushed 3T MRI acquisition resolutions below 1.0 mm for improved structure definition and morphometry. Yet, only few, time-intensive automated image analysis pipelines have been validated for high-resolution (HiRes) settings. Efficient deep learning approaches, on the other hand, rarely support more than one fixed resolution (usually 1.0 mm). Furthermore, the lack of a standard submillimeter resolution as well as limited availability of diverse HiRes data with sufficient coverage of scanner, age, diseases, or genetic variance poses additional, unsolved challenges for training HiRes networks. Incorporating resolution-independence into deep learning-based segmentation, i.e., the ability to segment images at their native resolution across a range of different voxel sizes, promises to overcome these challenges, yet no such approach currently exists. We now fill this gap by introducing a Voxel-size Independent Neural Network (VINN) for resolution-independent segmentation tasks and present FastSurferVINN, which (i) establishes and implements resolution-independence for deep learning as the first method simultaneously supporting 0.7–1.0 mm whole brain segmentation, (ii) significantly outperforms state-of-the-art methods across resolutions, and (iii) mitigates the data imbalance problem present in HiRes datasets. Overall, internal resolution-independence mutually benefits both HiRes and 1.0 mm MRI segmentation. With our rigorously validated FastSurferVINN we distribute a rapid tool for morphometric neuroimage analysis. The VINN architecture, furthermore, represents an efficient resolution-independent segmentation method for wider application.

Grapevine Segmentation in RGB Images using Deep Learning

Wine is the most important product from the Douro Region, in Portugal. Ampelographs are disappearing, and farmers need new solutions to identify grapevine varieties to ensure high-quality standards. The development of methodology capable of automatically identify grapevine are in need. In the scenario, deep learning based methods are emerging as the state-of-art in grapevines classification tasks. In previous work, we verify the deep learning models would benefit from focus classification patches in leaves images areas. Deep learning segmentation methods can be used to find grapevine leaves areas.This paper presents a methodology to segment grapevines images automatically based on the U-net model. A private dataset was used, composed of 733 grapevines images frames extracted from 236 videos collected in a natural environment. The trained model obtained a Dice of 95.6% and an Intersection over Union of 91.6%, results that fully satisfy the need of localise grapevine leaves.

Remote Sensing Plateau Forest Segmentation with Boundary Preserving Double Loss Function Collaborative Learning

Plateau forest plays an important role in the high-altitude ecosystem, and contributes to the global carbon cycle. Plateau forest monitoring request in-suit data from field investigation. With recent development of the remote sensing technic, large-scale satellite data become available for surface monitoring. Due to the various information contained in the remote sensing data, obtain accurate plateau forest segmentation from the remote sensing imagery still remain challenges. Recent developed deep learning (DL) models such as deep convolutional neural network (CNN) has been widely used in image processing tasks, and shows possibility for remote sensing segmentation. However, due to the unique characteristics and growing environment of the plateau forest, generate feature with high robustness needs to design structures with high robustness. Aiming at the problem that the existing deep learning segmentation methods are difficult to generate the accurate boundary of the plateau forest within the satellite imagery, we propose a method of using boundary feature maps for collaborative learning. There are three improvements in this article. First, design a multi input model for plateau forest segmentation, including the boundary feature map as an additional input label to increase the amount of information at the input. Second, we apply a strong boundary search algorithm to obtain boundary value, and propose a boundary value loss function. Third, improve the Unet segmentation network and combine dense block to improve the feature reuse ability and reduces the image information loss of the model during training. We then demonstrate the utility of our method by detecting plateau forest regions from ZY-3 satellite regarding to Sanjiangyuan nature reserve. The experimental results show that the proposed method can utilize multiple feature information comprehensively which is beneficial to extracting information from boundary, and the detection accuracy is generally higher than several state-of-art algorithms. As a result of this investigation, the study will contribute in several ways to our understanding of DL for region detection and will provide a basis for further researches.

Robust deep 3-D architectures based on vascular patterns for liver vessel segmentation

Deep Learning is providing new solutions for medical image segmentation problems and becomes a key point for future clinical application. The study of vascular structures is a challenging task due to the extremely small size of the vessel structure, low SNR, and varying contrast in medical image data.In this study, we present an end-to-end deep learning segmentation method relying on the integration of vessel enhancement filters inside a 3-D U-Net based architecture. In particular, the raw data used in the learning process -or used as input- is preprocessed using these filters. The use of vesselness filters can significantly improve the contrast in raw images; this step can also help to improve segmentation decision especially on bifurcations. 3-D U-Net, Dense U-Net and MultiRes U-Net are pitted against each other in the vessel segmentation task with the public IRCAD dataset. Considering the integration of vesselness filters, the model parameters were optimized in order to identify the optimal configuration for fully automatic segmentation of hepatic vessels. In addition the three architectures were tested on full 3-D images and slabs (stacks of 2-D slices). The results showed that the most accurate setup is the full 3-D process, which provides the highest Dice for most of the considered models. The 3-D Dense U-Net also gives the best results compared to other models with or without vessel enhancement filters.

State-of-the-Art Challenges and Perspectives in Multi-Organ Cancer Diagnosis via Deep Learning-Based Methods.

Simple SummaryCancer is a deadly disease that needs to be diagnose at early stage to increase patient survival rate. Multi-organ (such as breast, brain, lung, and skin) cancer detection, segmentation and classification manually using medical imaging is time consuming and required high expertise. In this study, we summarize existing deep learning segmentation and classification methods for multi-organ cancer diagnosis and provide future challenges with possible solutions. This review may benefit researchers to design new robust approaches that could be useful for the medical specialists as a second view.Thus far, the most common cause of death in the world is cancer. It consists of abnormally expanding areas that are threatening to human survival. Hence, the timely detection of cancer is important to expanding the survival rate of patients. In this survey, we analyze the state-of-the-art approaches for multi-organ cancer detection, segmentation, and classification. This article promptly reviews the present-day works in the breast, brain, lung, and skin cancer domain. Afterwards, we analytically compared the existing approaches to provide insight into the ongoing trends and future challenges. This review also provides an objective description of widely employed imaging techniques, imaging modality, gold standard database, and related literature on each cancer in 2016–2021. The main goal is to systematically examine the cancer diagnosis systems for multi-organs of the human body as mentioned. Our critical survey analysis reveals that greater than 70% of deep learning researchers attain promising results with CNN-based approaches for the early diagnosis of multi-organ cancer. This survey includes the extensive discussion part along with current research challenges, possible solutions, and prospects. This research will endow novice researchers with valuable information to deepen their knowledge and also provide the room to develop new robust computer-aid diagnosis systems, which assist health professionals in bridging the gap between rapid diagnosis and treatment planning for cancer patients.

A Denoising Self-supervised Approach for COVID-19 Pneumonia Lesion Segmentation with Limited Annotated CT Images.

The coronavirus disease 2019 (COVID-19) has become a global pandemic. The segmentation of COVID-19 pneumonia lesions from CT images is important in quantitative evaluation and assessment of the infection. Though many deep learning segmentation methods have been proposed, the performance is limited when pixel-level annotations are hard to obtain. In order to alleviate the performance limitation brought by the lack of pixel-level annotation in COVID-19 pneumonia lesion segmentation task, we construct a denoising self-supervised framework, which is composed of a pretext denoising task and a downstream segmentation task. Through the pretext denoising task, the semantic features from massive unlabelled data are learned in an unsupervised manner, so as to provide additional supervisory signal for the downstream segmentation task. Experimental results showed that our method can effectively leverage unlabelled images to improve the segmentation performance, and outperformed reconstruction-based self-supervised learning when only a small set of training images are annotated.Clinical relevance-The proposed method can effectively leverage unlabelled images to improve the performance for COVID-19 pneumonia lesion segmentation when only a small set of CT images are annotated.

Evaluation of Commercial Pelvic Autosegmentation Solutions in Anatomic Edge Cases of Patients Receiving Definitive Radiotherapy for Prostate Cancer

To assess the performance of commercially available autosegmentation solutions of pelvic anatomy in edge cases of anatomic variation among patients receiving definitive radiotherapy (RT) for prostate cancer.In a single institution retrospective cohort of 830 patients with prostate cancer receiving definitive RT between 2011 and 2019, we identified 112 patients with anatomic variations seen on simulation CT imaging which may present challenging cases for precise delineation of anatomy, including reasons such as arthroplasty metal artifacts, extensive median lobe hypertrophy, "droopy" seminal vesicles, presence of urinary catheter, and history of transurethral resection of prostatic tissue. Three commercially available solutions for pelvic anatomy autosegmentation (AS) employing deep learning and model-based segmentation were applied to treatment planning CT scans to generate segmented volumes of the prostate, rectum, bladder and femoral heads in DICOM-RT format. To quantify the accuracy of software-generated contours, Dice similarity coefficients (DSC) and Hausdorff distances (HD) were calculated for each AS contour with a comparator manually segmented reference contour. Calculations were performed in R using the RadOnc package and manually segmented contours were approved by an expert radiation oncologist with 19 years of experience.A total of 112 patients were identified for inclusion after manual review of treatment planning CTs. Contours were generated using two atlas/model-based segmentation products and a deep learning segmentation method. The mean DSC for the prostate, bladder, rectum was calculated for each segmentation method (Table 1). Deep learning segmentation outperformed model-based methods for all structures with the highest mean DSC but still had significant disagreement with manually segmented structures. Hip arthroplasty in particular reduced overall performance more than other anatomical edge cases, followed by prostatic hypertrophy.Anatomic edge cases present a challenging and relevant consideration in clinical implementation of autosegmentation software.

A Robust Deep Learning Segmentation Method for Hematoma Volumetric Detection in Intracerebral Hemorrhage.

Hematoma volume (HV) is a significant diagnosis for determining the clinical stage and therapeutic approach for intracerebral hemorrhage (ICH). The aim of this study is to develop a robust deep learning segmentation method for the fast and accurate HV analysis using computed tomography. A novel dimension reduction UNet (DR-UNet) model was developed for computed tomography image segmentation and HV measurement. Two data sets, 512 ICH patients with 12 568 computed tomography slices in the retrospective data set and 50 ICH patients with 1257 slices in the prospective data set, were used for network training, validation, and internal and external testing. Moreover, 13 irregular hematoma cases, 11 subdural and epidural hematoma cases, and 50 different HV cases into 3 groups (<30, 30-60, and >60 mL) were selected to further evaluate the robustness of DR-UNet. The image segmentation performance of DR-UNet was compared with those of UNet, the fuzzy clustering method, and the active contour method. The HV measurement performance was compared using DR-UNet, UNet, and the Coniglobus formula method. Using DR-UNet, the segmentation model achieved a performance similar to that of expert clinicians in 2 independent test data sets containing internal testing data (Dice of 0.861±0.139) and external testing data (Dice of 0.874±0.130). The HV measurement derived from DR-UNet was strongly correlated with that from manual segmentation (R2=0.9979; P<0.0001). In the irregularly shaped hematoma group and the subdural and epidural hematoma group, DR-UNet was more robust than UNet in both hematoma segmentation and HV measurement. There is no statistical significance in segmentation accuracy among 3 different HV groups. DR-UNet can segment hematomas from the computed tomography scans of ICH patients and quantify the HV with better accuracy and greater efficiency than the main existing methods and with similar performance to expert clinicians. Due to robust performance and stable segmentation on different ICHs, DR-UNet could facilitate the development of deep learning systems for a variety of clinical applications.