Automatic segmentation and recognition of pathomorphological microscopic images

We have been developing a computer-aided diagnosis (CAD) system for automatically recognizing cervical cancer cells from Papanicolaou smear. Considering that pathological changes of cervix can be indicated by the abnormity of the nucleus of intermediate cell, the key task of this system is to find the intermediate cells and segment the nucleus precisely. This paper presents a novel approach for automatic segmentation of microscopic cervical cell images using multispectral imaging techniques. In order to capture images at different wavelengths, a Liquid Crystal Tunable Filter (LCTF) device is used to provide wavelength selection from 400nm to 720nm with an increment of 10nm. Considering the spectral variances of background, nucleus and cytoplasm, background is extracted firstly from the microscopic images by calculating pixel intensity variance at 470nm, 530nm, 570nm, 580nm and 650nm. Then superficial cells are extracted apart from intermediate cells easily at 530nm 650nm because of the different pixel intensity distribution of the two kinds of cells at these two wavelengths. To segment the nucleus from intermediate cells, we adopt two procedures. Firstly, the nuclei are roughly segmented apart by using an iterative maximum deviation between-cluster algorithm. Secondly, a novel rigorous algorithm based on active contour model is adopted to achieve more exact nuclei segmentation. Using the method proposed in this paper, we did experiments on over 300 cervical smears, and the results show that this method is more robust and precise.

To construct deep learning-assisted diagnosis models based on automatic segmentation of ultrasound images to facilitate radiologists in differentiating benign and malignant parotid tumors. A total of 582 patients histopathologically diagnosed with PGTs were retrospectively recruited from 4 centers, and their data were collected for analysis. The radiomics features of six deep learning models (ResNet18, Inception_v3 etc) were analyzed based on the ultrasound images that were obtained under the best automatic segmentation model (Deeplabv3, UNet++, and UNet). The performance of three physicians was compared when the optimal model was used and not. The Net Reclassification Index (NRI) and Integrated Discrimination Improvement (IDI) were utilized to evaluate the clinical benefit of the optimal model. The Deeplabv3 model performed optimally in terms of automatic segmentation. The ResNet18 deep learning model had the best prediction performance, with an area under the receiver-operating characteristic curve of 0.808 (0.694-0.923), 0.809 (0.712-0.906), and 0.812 (0.680-0.944) in the internal test set and external test sets 1 and 2, respectively. Meanwhile, the optimal model-assisted clinical and overall benefits were markedly enhanced for two out of three radiologists (in internal validation set, NRI: 0.259 and 0.213 [p = 0.002 and 0.017], IDI: 0.284 and 0.201 [p = 0.005 and 0.043], respectively; in external test set 1, NRI: 0.183 and 0.161 [p = 0.019 and 0.008], IDI: 0.205 and 0.184 [p = 0.031 and 0.045], respectively; in external test set 2, NRI: 0.297 and 0.297 [p = 0.038 and 0.047], IDI: 0.332 and 0.294 [p = 0.031 and 0.041], respectively). The deep learning model constructed for automatic segmentation of ultrasound images can improve the diagnostic performance of radiologists for PGTs.

Object density-based image segmentation and its applications in biomedical image analysis

Tumor computed tomography (CT) image segmentation can provide a basis for the determination of tumor location and type. Therefore, it is of great significance to study the automatic segmentation method of tumor CT images. To address the problem of poor segmentation effect of traditional automatic tumor CT images segmentation methods, we propose an automatic segmentation method for tumor CT images using deep convolutional neural networks (DCNNs). First, the CT tumor image is simplified. According to the features of the target region and the background region, the distribution features of the tumor region in the CT images are obtained by convolution calculation, and the feature extraction is completed by feature fusion. Second, based on the feature extraction results, a deep supervised network is constructed to determine the image depth, which lays a solid foundation for accurate segmentation of tumor regions. Finally, DCNN was used to construct automatic segmentation for tumor CT images, which achieves the automatic segmentation of tumor CT images by mode calculation. The results show that the segmented tumor region is close to the actual region and the maximum pixel loss coefficient is 0.07, the maximum segmentation sensitivity is 7865[Formula: see text]kbps/s, the pixel segmentation specific coefficient and the segmentation edge distance are kept at a low level, which has a certain application value in the field of tumor CT images.

Artificial intelligence algorithms have been used in a wide range of applications in clinical aided diagnosis, such as automatic MR image segmentation and seizure EEG signal analyses. In recent years, many machine learning-based automatic MR brain image segmentation methods have been proposed as auxiliary methods of medical image analysis in clinical treatment. Nevertheless, many problems regarding precise medical images, which cannot be effectively utilized to improve partition performance, remain to be solved. Due to the poor contrast in grayscale images, the ambiguity and complexity of MR images, and individual variability, the performance of classic algorithms in medical image segmentation still needs improvement. In this paper, we introduce a distributed multitask fuzzy c-means (MT-FCM) clustering algorithm for MR brain image segmentation that can extract knowledge common among different clustering tasks. The proposed distributed MT-FCM algorithm can effectively exploit information common among different but related MR brain image segmentation tasks and can avoid the negative effects caused by noisy data that exist in some MR images. Experimental results on clinical MR brain images demonstrate that the distributed MT-FCM method demonstrates more desirable performance than the classic signal task method.

Study question Could hyperspectroscopy (HS) allow for the identification of a functionally apt spermatozoon once it is immobilized before its microinjection into the oocyte? Summary answer Yes, HS coupled to machine learning (ML) can predict the success of a single sperm in developing into a high-quality embryo from its hyperspectral profile/signature. What is known already HS is a non-contact, non-ionizing imaging technique combining the spatial information provided by optical microscope images and the chemical information recovered from the emission spectra recorded at a vast array of wavelengths. HS has been used in other medical fields to identify patterns in the hyperspectral signature of specific cell types such as cancerous versus non-cancerous tissue – to define them and categorize them in an immediate non-invasive way. It has been previously studied in reproductive science to predict the quality and success of mouse oocytes and embryos based on the relative abundance of metabolites or distribution of organelles. Study design, size, duration Prospective interventional non-selection study. 76 patients undergoing ICSI with autologous fresh ejaculate sperm were recruited. Hyperspectral images were taken of 626 immobilized spermatozoa of which 552 were used to feed different iterations of classic ML predictive models based on XGBoost decision trees. All the remaining steps of the ICSI cycle were performed according to routine laboratory procedure at IVIRMA Valencia. The study lasted from January 2022 to May 2024. Participants/materials, setting, methods Glass-bottom dishes were used for the ICSI procedure. Images were acquired using a 4200M Microscope system coupled to an IX73 inverted microscope with micromanipulator systems, and TruScope v.1.17 software. Automatic segmentation of the images was performed using a U-Net type convolutional neural network (CNN). Clinical and hyperspectral data were pre-processed and then fed to different iterations of ML models in Python 3.11.2. Performance metrics such as precision, sensitivity and specificity were evaluated. Main results and the role of chance The automatic segmentation of hyperspectral images was proven successful, resulting in an average Dice coefficient of 88.18%. Different iterations of ML XGBoost models were developed using either hypercubes as input or both hypercubes and clinical data. Chained and independent models were used to predict the three outcomes studied: fertilization, arrival to blastocyst and usable blastocyst (of sufficient quality to transfer fresh or cryopreserve). The precision of the prediction for each imaged sperm was compared to the success achieved by the embryologists in the study’s ICSI cycles: 79.9% (95% confidence interval (CI): 73.0-87.2) fertilization, 59.6% (53.7-66.0) blastulation, 37.9% (33.2-43.0) usable blastocyst rate. The most promising iterations were those which had an appropriate balance between high precision and a narrow 95% CI. Independent models using both clinical and hyperspectral data resulted in an 83.7% (76.2-89.2) precision predicting fertilization, 63.5% (52.1-73.6) blastulation and 59.2% (41.1-69.6) usable blastocyst. Chained models using only hyperspectral data resulted in 85.3% (76.8-91.0) precision in prediction of fertilization, 72.4% (59.8-82.2) blastulation and 78.9% (56.7-91.5) usable blastocyst. This last iteration provided a higher precision in the prediction of the arrival to a usable blastocyst than the result obtained by the embryologists, without overlapping 95% CI. Limitations, reasons for caution The limited number of valid input images as well as the class imbalance hindered results, specifically by widening the uncertainty in the prediction for each sperm imaged. The study relies on the procedures and results of a single lab which somewhat limits the generalizability of the proposed tool. Wider implications of the findings One iteration of the proposed model obtained significantly higher precision in predicting the ability of a sperm to fertilize and develop into a usable blastocyst than the embryologists’ selection during standard ICSI. Thus, once optimized, HS could aid sperm selection during ICSI associating each sperm to a specific success rate. Trial registration number No

Since the introduction of 3-D rotational X-ray imaging, protocols for 3-D rotational coronary artery imaging have become widely available in routine clinical practice. Intra-procedural cardiac imaging in a computed tomography (CT)-like fashion has been particularly compelling due to the reduction of clinical overhead and ability to characterize anatomy at the time of intervention. We previously introduced a clinically feasible approach for imaging the left atrium and pulmonary veins (LAPVs) with short contrast bolus injections and scan times of approximately 4 -10 s. The resulting data have sufficient image quality for intra-procedural use during electro-anatomic mapping (EAM) and interventional guidance in atrial fibrillation (AF) ablation procedures. In this paper, we present a novel technique to intra-procedural surface generation which integrates fully-automated segmentation of the LAPVs for guidance in AF ablation interventions. Contrast-enhanced rotational X-ray angiography (3-D RA) acquisitions in combination with filtered-back-projection-based reconstruction allows for volumetric interrogation of LAPV anatomy in near-real-time. An automatic model-based segmentation algorithm allows for fast and accurate LAPV mesh generation despite the challenges posed by image quality; relative to pre-procedural cardiac CT/MR, 3-D RA images suffer from more artifacts and reduced signal-to-noise. We validate our integrated method by comparing 1) automatic and manual segmentations of intra-procedural 3-D RA data, 2) automatic segmentations of intra-procedural 3-D RA and pre-procedural CT/MR data, and 3) intra-procedural EAM point cloud data with automatic segmentations of 3-D RA and CT/MR data. Our validation results for automatically segmented intra-procedural 3-D RA data show average segmentation errors of 1) approximately 1.3 mm compared with manual 3-D RA segmentations 2) approximately 2.3 mm compared with automatic segmentation of pre-procedural CT/MR data and 3) approximately 2.1 mm compared with registered intra-procedural EAM point clouds. The overall experiments indicate that LAPV surfaces can be automatically segmented intra-procedurally from 3-D RA data with comparable quality relative to meshes derived from pre-procedural CT/MR.

Cardiac MRI has experienced a crescent relevance in clinical investigations. The segmentation of myocardial walls is a prerequisite for assessment of cardiac viability. Manual or semi-automatic segmentation of all the images of a subject is tedious, as well as consuming much time from cardiologists. In this study, we selected 23 slices of simulated cardiac MR by MRXCAT and 30 real slices of CINE-MR from 15 patients with Chagas Disease. The proposed pipeline of the fully automatic segmentation consists of three steps: 1. Preprocessing; 2. Automatic Seeds Definition; and 3. Segmentation by Geodesic Active Contour. An experienced cardiologist provided the gold standard annotations of apical, mid-ventricular and basal LV myocardium. We use the following three metrics to validate the proposed pipeline with different signal to noise ratio: Dice similarity (DS), Precision (Pr) and Volumetric Similarity (VS). DS show good agreement between manual segmentation and the automatic segmentation in simulated images with SNR 200, 25, 15 and 5, i.e., 0.98, 0.93, 0.9 and 0.72, respectively. We found moderate agreements between manual segmentation and Snake segmentation in simulated images with SNR 200, 25, 15 and 5, i.e., 0.38, 0.42, 0.34 and 0.39, respectively. The DS, VS, and Pr obtained suggest substantial agreements between the manual and our proposed method segmentation in images of Chagas’s Disease, i.e., 0.8 [0.69–0.87], 0.89 [0.72–0.99] and 0.9 [0.76–0.98] (mean [min–max]), respectively. Our findings suggest that one can use the proposed method in the automatic myocardium segmentation with reliability similar to manual tracing, although completely free of human interaction.

Pulmonary lobes segmentation of pulmonary CT images is important for assistant therapy and diagnosis of pulmonary disease in many clinical tasks. Recently supervised deep learning methods are applied widely in fast automatic medical image segmentation including pulmonary lobes segmentation of pulmonary CT images. However, they require plenty of ground truth due to their supervised learning scheme, which are always difficult to realize in practice. To address this issue, in this study we extend an existed unsupervised learning network with an extra pulmonary mask constraint to develop a deformable pulmonary lobes atlas and apply it for fast automatic segmentation of pulmonary lobes in pulmonary CT images. The experiment on 40 pulmonary CT images shows that our method can segment the pulmonary lobes in seconds, and achieve average Dice of 0.906 ± 0.044 and average surface distance of 0.495 ± 0.380 mm, which outperforms the state-of-the-art methods in segmentation accuracy. Our method successfully combines the advantages of both deformable atlas and unsupervised learning for automatic segmentation and ensures the consistent and topology preserving of pulmonary lobes without any postprocessing.

As is known, pathologists visually examine the tissue distributions by using microscopes traditionally. The rise in digital image processing and machine learning also allows high-performance computerized analysis of histology images taken with modern imaging systems. In general, histological image segmentation is the first step in the quantitative analysis of histology images. Therefore, a high-accuracy segmentation is essential for histology image analysis in most cases. In this paper, we performed a deep Convolutional Neural Networks (CNNs) based nuclei segmentation study on colon histology images. By using the U-Net biomedical image segmentation model, it is aimed to classify each pixel in colon histology images into one of the following 6 types of nucleus: epithelial, lymphocyte, plasma, eosinophil, neutrophil, connective tissue or classify it as the image background. In comprehensive experimental tests performed on Colon Nuclei Identification and Counting (CoNIC) Challenge dataset, commonly used segmentation and classification metrics were measured, and promising segmentation performances were achieved.

Elastography is a new ultrasonic method for measuring tissuespsila elasticity. Besides many and widely acknowledged benefits, the method suffers severe limitations due to the high motion sensitivity and inter-operator dependency to the point where it provides only qualitative information, not having, until now, any real quantification means. In this paper we present an automatic segmentation method for elastographic images based on statistical techniques. First a probabilistic model is built for every pixel in the image, derived by processing a video sequence instead of a single image. The built image contains the values with highest probability for each pixel. Next a DAEM (Deterministic Annealing Expectation Maximization) method is used for automatic image segmentation. Finally a numerical quantification of tissue elasticity is provided based on the segmentation.

This paper describes the application and validation of automatic segmentation of three-dimensional images by non-rigid registration to atlas images. The registration-based segmentation technique is applied to confocal microscopy images acquired from the brains of 20 bees. Each microscopy image is registered to an already segmented reference atlas image using an intensity-based non-rigid image registration algorithm. This paper evaluates and compares four different approaches: registration to an individual atlas image (IND), registration to an average shape atlas image (AVG), registration to the most similar image from a database of individual atlas images (SIM), and registration to all images from a database of individual atlas images with subsequent fuzzy segmentation (FUZ). For each strategy, the segmentation performance of the algorithm was quantified using both a global segmentation correctness measure and the similarity index. Manual segmentation of all microscopy images served as a gold standard. The best segmentation result (median correctness 91 percent of all voxels) was achieved using the FUZ paradigm. Robustness was also the best for this strategy (minimum correctness over all individuals 84 percent). The mean similarity index value of segmentations produced by the FUZ paradigm is 0.86 (IND, 0.81; AVG, 0.84; SIM, 0.82). The superiority of the FUZ paradigm is statistically significant (two-sided paired t-test, P<0.001).

Event Abstract Back to Event Study of cerebral asymmetry in schizophrenia based on an novel automatic cerebral hemisphere segmentation method for MRI Lu Zhao1*, Ulla Ruotsalainen1, Jussi Hirvonen2, Jarmo Hietala3 and Jussi Tohka1 1 Tampere University of Technology, Finland 2 Turku PET Centre, Finland 3 University of Turku, Finland In this work, we statistically analyzed the cerebral volume asymmetry for schizophrenic and healthy populations by employing a novel and completely automatic cerebral hemisphere segmentation method for MRI. The segmentation method is based on the information of partial volume effect (PVE) and the partial differential equations based shape bottlenecks (PDE-SB) algorithm [1]. PVE is an inevitable problem of MRI, but it is rarely taken into account by the existing automatic MR brain image segmentation techniques. The PDE-SB algorithm can detect the bottlenecks connecting two parts of a complex object by simulating a steady state of the information transmission process between the two parts with partial differential equations. Our method consists of two major stages: compartmental brain decomposition [2] and cerebral hemisphere segmentation [3]. Firstly, a partial volume estimation approach [4] is used to obtain the partial volume voxel classification [into grey matter (GM), white matter (WM), cerebrospinal fluid (CSF), CSF/GM, GM/WM and CSF/background] and estimate the amount of each tissue type in each brain voxel. Then the WM+GM/WM region is split into compartmental seeds corresponding to cerebrum, cerebellum and brainstem utilizing the PDE-SB algorithm. The brain volume is compartmentally decomposed by reconstructing the original shapes of the three compartments from the compartmental seeds according to the compartmental boundary knowledge defined with partial volume information. The cerebrum portion is extracted from the decomposed brain volume, and the cerebral hemispheres are segmented with the PDE-SB algorithm and the partial volume estimation results. In this work, this method was applied to a set of clinical T1-weighted MR images of 18 schizophrenic (11 males, 7 females) and 19 healthy (12 males, 7 females) subjects. All subjects were right-handed, and the schizophrenic patients were never medicated. With the obtained PVE information and cerebral hemisphere segmentation, the asymmetry index (AI) was calculated for cerebrum, cerebral GM and cerebral WM. AI was defined as (VR-VL)/[1/2(VR+VL)]×100, where VR and VL were the volume sizes of the right and left hemispheres respectively. The permutation test was conducted to evaluate the statistical significance of the difference of acquired AIs between schizophrenic and healthy populations. Here, the null hypothesis was defined as that AIs of schizophrenic and healthy populations had identical probability distribution, and the significance level was selected as 0.05. According to the statistical analysis results, the cerebral GM volume asymmetry of female schizophrenic population had significantly different probability distribution from the female healthy population's (p=0.002). There were no significant difference between schizophrenic and healthy populations with regard to volume asymmetry in the other cases (p>0.05). Conference: Neuroinformatics 2008, Stockholm, Sweden, 7 Sep - 9 Sep, 2008. Presentation Type: Poster Presentation Topic: Neuroimaging Citation: Zhao L, Ruotsalainen U, Hirvonen J, Hietala J and Tohka J (2008). Study of cerebral asymmetry in schizophrenia based on an novel automatic cerebral hemisphere segmentation method for MRI. Front. Neuroinform. Conference Abstract: Neuroinformatics 2008. doi: 10.3389/conf.neuro.11.2008.01.069 Copyright: The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers. They are made available through the Frontiers publishing platform as a service to conference organizers and presenters. The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated. Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed. For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions. Received: 28 Jul 2008; Published Online: 28 Jul 2008. * Correspondence: Lu Zhao, Tampere University of Technology, Tampere, Finland, lu.zhao@tut.fi Login Required This action requires you to be registered with Frontiers and logged in. To register or login click here. Abstract Info Abstract The Authors in Frontiers Lu Zhao Ulla Ruotsalainen Jussi Hirvonen Jarmo Hietala Jussi Tohka Google Lu Zhao Ulla Ruotsalainen Jussi Hirvonen Jarmo Hietala Jussi Tohka Google Scholar Lu Zhao Ulla Ruotsalainen Jussi Hirvonen Jarmo Hietala Jussi Tohka PubMed Lu Zhao Ulla Ruotsalainen Jussi Hirvonen Jarmo Hietala Jussi Tohka Related Article in Frontiers Google Scholar PubMed Abstract Close Back to top Javascript is disabled. Please enable Javascript in your browser settings in order to see all the content on this page.

The morphological analysis of dendritic spines is an important challenge for the neuroscientific community. Most state-of-the-art techniques rely on user-supervised algorithms to segment the spine surface, especially those designed for light microscopy images. Therefore, processing large dendritic branches is costly and time-consuming. Although deep learning (DL) models have become one of the most commonly used tools in image segmentation, they have not yet been successfully applied to this problem. In this article, we study the feasibility of using DL models to automatize spine segmentation from confocal microscopy images. Supervised learning is the most frequently used method for training DL models. This approach requires large data sets of high-quality segmented images (ground truth). As mentioned above, the segmentation of microscopy images is time-consuming and, therefore, in most cases, neuroanatomists only reconstruct relevant branches of the stack. Additionally, some parts of the dendritic shaft and spines are not segmented due to dyeing problems. In the context of this research, we tested the most successful architectures in the DL biomedical segmentation field. To build the ground truth, we used a large and high-quality data set, according to standards in the field. Nevertheless, this data set is not sufficient to train convolutional neural networks for accurate reconstructions. Therefore, we implemented an automatic preprocessing step and several training strategies to deal with the problems mentioned above. As shown by our results, our system produces a high-quality segmentation in most cases. Finally, we integrated several postprocessing user-supervised algorithms in a graphical user interface application to correct any possible artifacts.

IntroductionSurgical treatment for subglottic stenosis often faces recurrence of narrowing and patients continue to face quality of life issues. Currently, biomechanical models that inform surgical treatments are difficult to create due to vocal fold intricacies and trouble distinguishing between tissue types–muscle, mucosa, and cartilage. Improved models of laryngeal microanatomy can be used to assess and predict surgical outcomes for patients who suffer from subglottic stenosis. While manual segmentation of histology is the standard, automatic segmentation can be more efficient, sensitive, and specific in tissue differentiation. Manual segmentation faces challenges like extensive operator time and inter‐rater subjectivity. The aim of this study, therefore, was to qualitatively compare manual segmentation to automatic k‐means segmentation of whole‐slide histology rabbit vocal fold images. This is the preliminary step towards enhancing segmentation and, later, achieving a high‐fidelity human larynx model.Methods134 whole‐slide histology images were manually segmented from seven rabbits using colored masks for muscle, cartilage, mucosa, and adipose tissue. Of those images, 31 were also selected for segmentation using k‐means for initial qualitative comparison. At least two images were segmented per rabbit. K‐means segmentation is optimal for our multidimensional data. Since it is deterministic, we selected the number of clusters as our parameter. After inspecting n=1‐10 clusters, n=4 was chosen as qualitatively producing the most distinct clusters without oversampling.ResultsFor our preliminary results, we compared the histological image, the manual segmentation, and the automatically segmented image. In the automatic segmentations, tissue type is distinguished by the grayscale contrast. The four classes visible are cartilage, muscle, adipose tissue, and mucosa. There is occasional loss of visualization of adipose tissue and mucosa, however, the cartilage and muscle tissues remain distinct. The mucosal area boundaries remain clear despite some visualization loss within the mucosal borders, therefore, mucosa can be distinguished from adjacent cartilage and muscle. Training on a large dataset like other machine learning algorithms is unnecessary for k‐means, which is ideal for our relatively small sample size. The speed and versatility of k‐means has qualitatively demonstrated comparable results to our manual segmentations.ConclusionFuture directions for our project include conducting k‐means segmentation of the remainder of the 134 images collected. We have also tested mean‐shift clustering and a deep learning convolutional neural network (CNN) to compare automatic segmentation methods to one another, and to the standard manual segmentation/ground truth. Future testing also includes quantitative analysis of these algorithms’ accuracy. Beyond our histology images, we have at least 14 MR images (at least two per rabbit) of the larynges as well. We intend to combine and overlay the structural detail of the histology images with the soft tissue differentiation of the MRI.