Segmented Regions Research Articles

Pathologic complete response prediction in breast cancer lesion segmentation and neoadjuvant therapy.

Predicting whether axillary lymph nodes could achieve pathologic Complete Response (pCR) after breast cancer patients receive neoadjuvant chemotherapy helps make a quick follow-up treatment plan. This paper presents a novel method to achieve this prediction with the most effective medical imaging method, Dynamic Contrast-enhanced Magnetic Resonance Imaging (DCE-MRI). In order to get an accurate prediction, we first proposed a two-step lesion segmentation method to extract the breast cancer lesion region from DCE-MRI images. With the segmented breast cancer lesion region, we then used a multi-modal fusion model to predict the probability of axillary lymph nodes achieving pCR. We collected 361 breast cancer samples from two hospitals to train and test the proposed segmentation model and the multi-modal fusion model. Both segmentation and prediction models obtained high accuracy. The results show that our method is effective in both the segmentation task and the pCR prediction task. It suggests that the presented methods, especially the multi-modal fusion model, can be used for the prediction of treatment response in breast cancer, given data from noninvasive methods only.

Cerebral amyloid angiopathy is associated with higher gray matter regional R2 relaxation rate: An ex‐vivo MRI and pathology study

AbstractBackgroundCerebral amyloid angiopathy (CAA) is characterized by accumulation of amyloid‐ß (Aß) protein in the walls of cortical and meningeal small vessels. CAA is common in older adults. It has been linked to intracerebral hemorrhage and cognitive decline. In this study, the association of CAA with the transverse relaxation rate R2 values was investigated in gray matter regions from a large number of community‐based older adults.MethodCerebral hemispheres from 802 older adults participating in four cohort studies of aging, the Rush Memory and Aging Project, the Minority Aging Research Study, the Religious Orders Study, and the Clinical Core of the Rush Alzheimer’s Disease Research Center (Fig. 1a) were included in this work. All hemispheres were imaged ex‐vivo using 3T MRI scanners. R2 maps were generated from multi‐echo spin‐echo images. Gray matter was segmented into 34 cortical and 8 subcortical regions. Median R2 values were obtained from each region. Following ex‐vivo MRI, all hemispheres underwent detailed neuropathologic assessment (Fig. 1b). Evaluated neuropathologies included: CAA, Aß plaques, neurofibrillary tangles, gross infarcts, microinfarcts, TDP‐43 pathology, hippocampal sclerosis, Lewy bodies, atherosclerosis, and arteriolosclerosis. Multiple linear regression was conducted in each segmented gray matter region separately to test the association of the corresponding median R2 value with the global CAA score, controlling for all other neuropathologies, age at death, sex, education, postmortem interval to fixation and to imaging, and scanners. False discovery rate (FDR) was used to correct for multiple comparisons. Statistical significance was set at p<0.05.ResultA significant positive association of median R2 with CAA pathology was found in several cortical regions of the temporal and frontal lobes as well as in subcortical structures (Fig. 2,3). No gray matter region showed lower median R2 for higher CAA score. Furthermore, we obtained similar findings when also controlling in the linear regression models for normalized regional volume and for clinical variables such as history of hypertension, diabetes, smoking, blood pressure levels, and the presence of the APOE‐e4 allele.ConclusionThis investigation showed that CAA is independently associated with elongated R2 in cortical regions of the temporal and frontal lobes as well as in subcortical structures.

Fully automatic deep‐learning based segmentation of Hippocampal subfields from standard resolution T1W MRI in Alzheimer’s Disease

AbstractBackgroundHippocampal subfield volumes are of interest in Alzheimer’s disease (AD) as sub‐regions are affected at different points in the disease trajectory and may provide increased sensitivity as biomarkers compared to the whole hippocampus alone. Subfield volume estimation typically requires a high‐resolution T2‐weighted MRI to ensure accurate anatomical parcellation; however, this increases scan times and may not be present in historical datasets. Automatic methods exist for subfield segmentation from mono‐modal standard resolution T1W MRI, with deep‐learning approaches providing the most promising results to date. Here we extend such work employing deep learning approaches to provide flexible hippocampal subfield segmentation of standard resolution (1mm3) T1W MRI.MethodWe trained a collection of deep‐learning segmentation models to perform binary segmentation of the CA1‐3, CA4/DG, Subiculum, and whole Hippocampus, using a publicly available dataset of 25 subjects with manually segmented region labels on T1 MRI (1mm3). Multilabel segmentations were generated from combination of segmentation outputs.ResultTotal hippocampal and subfield multi‐label segmentations were generated for all test subjects in each cross validated loop and DICE overlap with the ground truth label computed. Preliminary results, computed using a 5‐fold cross validation, report mean DICE in whole Hippocampus = 0.904(std: 0.014), CA1‐3 = 0.842(std:0.019), CA4/DG = 0.767(std:0.04), and Subiculum = 0.791(std:0.033).Segmentation models were applied to 947 subjects with standard T1 MRI from the ADNI cohort and either a CN or MCI label. Subjects were selected with pre‐existing hippocampal subfield volumes, provided by ADNI, computed with the ASHS method using a multi‐modal T1 and hi‐res T2 MRI image. While the proposed method and ASHS employ subtly different parcellation schema we note significant correlation across similar regions between methods.Logistic regression of ICV corrected volumes and MMSE, correcting for age and sex, reveal both ASHS and the proposed method can significantly discriminate between CN and MCI groups (Table 1), with subfields areas providing greater discrimination that of the whole hippocampus.ConclusionHere we present preliminary results for a flexible deep‐learning approach to segmentation of hippocampal subfields from standard resolution T1 MRI alone; indicating comparable performance for group discrimination to an established multi‐modal standard.

Segmentation of hard exudate lesions in color fundus image using two-stage CNN-based methods

The presence of hard exudate (EX) lesions is an early clinical symptom of Diabetic Retinopathy (DR); its accurate segmentation is essential for diagnosis and treatment. Automatic segmentation of EX lesions is challenging because they have a wide range of sizes, shapes, and brightness, and can be confused with soft exudate lesions. In this paper, we present and assess an efficient segmentation method for interactively segmenting EX lesions in fundus images from DR patients. Our approach consisted of two main stages: (1) the first stage generates an automatic segmentation using a CNN-based method. (2) If the obtained segmentation is suboptimal, the second stage is used to refine each inaccurately segmented region using a CNN-based interactive segmentation method. To train and evaluate method performance, we used two public datasets (IDRiD and DDR) and two local datasets from two medical centers (EHos and EWT) containing a wide range of EX lesions. Several state-of-the art CNN-based models and the proposed method were trained and evaluated on IDRiD, DDR and EHos datasets using the area under the precision–recall curve (AUPR), the dice similarity coefficient (Dice) and the intersection over union (IoU). In addition, EWT dataset was used to evaluate the methods without retraining the CNN models. The evaluation results using IDRiD dataset showed that the proposed method achieved AUPR, mean Dice and mean IoU scores of 0.893, 76.6% and 62.4%, respectively, which are comparable to the state-of-the-art method on EX segmentation. When evaluated on the DDR dataset with large portion of small EX lesions, the proposed method achieved AUPR, mean Dice and mean IoU scores of 0.693, 60.8% and 45.5%, yielding state-of-the-art performance. Furthermore, with the main evaluation metric being AUPR, the evaluation results on two datasets, EHos and ETW, with scores of 0.689 and 0.667 demonstrated superiority of the proposed method over the other segmentation methods. The proposed two-stage CNN-based segmentation method is an useful approach for accurately segmenting EX lesions in color fundus images, enabling lesion quantification to aid experts in treating patients with DR. The source code, the trained CNN models and the data are publicly available at: https://github.com/dvquang2000/Interactive_HardExudates_Segmentation.

Solar-induced chlorophyll fluorescence extraction based on heterogeneous light distribution for improving in-situ chlorophyll content estimation

Accurate estimation of leaf chlorophyll content (LCC) in field for wheat crops is important to provide guidance for precision management. Solar-induced chlorophyll fluorescence (SIF) extraction based on hyperspectral imaging is an effective tool for LCC estimation, which is because SIF is a signal directly and intrinsically related to photosynthetic activity of crops. However, soil background and structural difference always lead to mixing of soil, shadowed and sunlit canopy components, and heterogeneous light distribution among canopy, which increase the complexity of SIF extraction and reduce the LCC estimation accuracy. The purpose of this study is to propose an accurate SIF extraction method by eliminating above challenges to improve the LCC estimation accuracy. Hyperspectral image and LCC measurements were conducted at the jointing and heading stages of wheat crops. Firstly, to eliminate the effect of soil background, two image segmentation methods, excess green index (ExG) and optimized soil-adjusted vegetation index (OSAVI), are compared to extract the crop region. Secondly, to reduce the canopy structure effect, light distribution condition is evaluated by the brightness, which is extracted from intensity component of hue-saturation-intensity color space. Then K-means clustering based on the brightness distribution is utilized to divide the segmented crop region further into shadowed and sunlit regions. Finally, SIF was extracted based on the Fraunhofer line discriminator method at 687 and 761 nm, and was modified by normalization of brightness to reduce the influence of heterogeneous light distribution. The performance of random forest regression models, which are built by extracting SIF based on raw image, segmented crop by ExG segmentation, segmented crop by OSAVI segmentation, and shadowed and sunlit regions, respectively, is compared. Results show the estimation accuracy of the models increases when image segmentation is performed (determination coefficient of prediction set (RP2) = 0.37 for ExG segmentation; RP2 = 0.47 for OSAVI segmentation) compared with raw image (RP2 = 0.30). Moreover, shadowed region clustering based on brightness was optimal region for SIF extraction (RP2 = 0.54) compared with the sunlit region (RP2 = 0.27). The modified SIF at shadowed regions, normalized by the brightness of crops to reduce the effect of heterogeneous light distribution, yields the highest relationship with LCC (RP2 = 0.79), which outperforms SIF normalized by the absorbed photosynthetically active radiation (APAR). The reason is that brightness is more related to LAI, a parameter representing canopy structure, than APAR. It suggests that OSAVI segmentation for soil background effect elimination and SIF extraction considering canopy structure and heterogeneous light distribution facilitate the improvement of LCC estimation accuracy for high-throughput phenotyping in field.

NIMG-42. RADIOIMMUNOMIC SIGNATURES IN PEDIATRIC LOW-GRADE GLIOMA FOR NON-INVASIVE IDENTIFICATION OF IMMUNOTHERAPEUTIC TARGETS

Abstract INTRODUCTION Tumor immune microenvironment (TIME) plays a key role in response to anti-tumor therapies. Therefore, detailed characterization of TIME is crucial for patient stratification and enrollment onto emerging immunotherapies. As standard-of-care treatments are not feasible or confer devastating long-term morbidity for pediatric low-grade glioma (pLGG), non-invasive quantification of the TIME using radiomic analysis, may guide treatment decisions. On the largest available transcriptomic and imaging data in pLGGs, provided through Children’s Brain Tumor Network (CBTN), this data-driven study aims to discover immunological profiles of pLGG that are meaningfully related to biomarkers of tumor immune response, and radioimmunomic signatures that are predictive of immunological profiles. METHODS Enrichment of immune and stromal cell types and pathways in 502 subjects were inferred using xCell algorithm applied to transcriptomic data and the tumors were then clustered via consensus clustering approach. Tumor inflammation signature (TIS), as a biomarker of response to anti-PD-1 blockade, was compared across the immunological clusters. Quantitative radiomic features were extracted from the segmented tumor regions on multiparametric MRI scans of 155/502 patients. Cross-validated support vector machines with forward feature selection were trained on radiomic features using one-versus-the-rest approach to predict the immunological groups. RESULTS Three immunological clusters, i.e., immune-cold, -altered, and -hot were found, with different levels of enrichment in immune and microenvironment scores, CD4 T-cells, and macrophages. Significantly higher TIS values were found in immune-hot and -altered compared to immune-cold tumors (p= 1.2e-8 and p= 7e-16, respectively). Radioimmunomic models yielded average AUC of 0.75 for one-versus-the-rest differentiation of the immunological clusters. CONCLUSIONS The proposed analysis revealed three distinct immunological groups in pLGGs, with immune-hot tumors more likely to respond favorably to immunotherapies. Furthermore, non-invasive radioimmunomic signatures were shown to potentially stratify the patients based on their TIME, which upon further development and validation, could inform treatments for pLGGs.

TMIC-49. AUTOMATIC SEGMENTATION OF DISTINCT GLIOBLASTOMA HALLMARKS ON DIGITIZED H&E SLIDES VIA DEEP LEARNING: CURATION OF A LARGE MULTI-INSTITUTIONAL COHORT

Abstract PURPOSE Glioblastoma (GBM) is a heterogeneous brain tumor. Standard treatment constitutes surgery, chemo-radiation, and chemotherapy. After surgery, the resected tissue is analyzed under a microscope by neuropathologists to identify distinct histological hallmarks. However, visual analysis of GBM pathological features on Hematoxylin and Eosin (H&E)-stained slides is subjective, time-consuming, and more importantly, requires wide expertise. We present a hierarchical deep learning-strategy to automatically segment distinct GBM hallmarks that first distinguishes necrosis (NE) from cellular tumor (CT), and subsequently, segments regions of microvascular proliferation (MV) and pseudo-palisading cell patterns (PC) within the CT region, on digitized H&E slides. METHODS We employed n=541 publicly available studies including IvyGap (n=41), and TCGA (n=500). For the IvyGap cohort, expert-driven segmentations of CT, NE, MV and PC were available. We randomly employed n=21 slides from 4 patients for training, n=5 slides from 2 cases for validation, and n=2 slides from 2 cases for testing. We employed an EfficientNet model using ~600 patches of 224×224 pixels of NE and CT sampled for every slide. For semantic segmentation of MV and PC, we separately trained two EfficientNet-UNet models trained on ~200 squared-patches of 1792 and 896 pixels size, respectively. RESULTS Our hierarchical deep learning model on the IvyGap cohort achieved an accuracy of 96%, with an area under the curve (AUC) of 98% for distinguishing necrosis from non-necrotic (i.e. CT) regions. For segmenting PC and MV within the non-necrotic regions, our DL models achieve a dice score of 0.69 and 0.77, respectively. Additionally, the segmentation results for each of the pathological attributes were also generated and vetted for the TCGA cohort. CONCLUSION We developed a hierarchical model for automatic segmentation of NE, CT, and MP and PC in H&E-stained Glioblastoma images which could reduce cost, time, and effort in annotating these pathological phenotypes, for quantifying the GBM micro-environment.

Machine Learning Pipeline for Predicting Bone Marrow Edema Along the Sacroiliac Joints on Magnetic Resonance Imaging.

We aimed to develop and validate a fully automated machine learning (ML) algorithm that predicts bone marrow edema (BME) on a quadrant level in sacroiliac (SI) joint magnetic resonance imaging (MRI). A computer vision workflow automatically locates the SI joints, segments regions of interest (ilium and sacrum), performs objective quadrant extraction, and predicts presence of BME, suggestive of inflammatory lesions, on a quadrant level in semicoronal slices of T1/T2-weighted MRI scans. Ground truth was determined by consensus among human readers. The inflammation classifier was trained using a ResNet18 backbone and five-fold cross-validated on scans of patients with spondyloarthritis (SpA) (n=279), postpartum individuals (n=71), and healthy subjects (n=114). Independent SpA patient MRI scans (n=243) served as test data set. Patient-level predictions were derived from aggregating quadrant-level predictions, ie, at least one positive quadrant. The algorithm automatically detects the SI joints with a precision of 98.4% and segments ilium/sacrum with an intersection over union of 85.6% and 67.9%, respectively. The inflammation classifier performed well in cross-validation: area under the curve (AUC) 94.5%, balanced accuracy (B-ACC) 80.5%, and F1 score 64.1%. In the test data set, AUC was 88.2%, B-ACC 72.1%, and F1 score 50.8%. On a patient level, the model achieved a B-ACC of 81.6% and 81.4% in the cross-validation and test data set, respectively. We propose a fully automated ML pipeline that enables objective and standardized evaluation of BME along the SI joints on MRI. This method has the potential to screen large numbers of patients with (suspected) SpA and is a step closer towards artificial intelligence-assisted diagnosis and follow-up.

Pixel Diffuser: Practical Interactive Medical Image Segmentation without Ground Truth.

Medical image segmentation is essential for doctors to diagnose diseases and manage patient status. While deep learning has demonstrated potential in addressing segmentation challenges within the medical domain, obtaining a substantial amount of data with accurate ground truth for training high-performance segmentation models is both time-consuming and demands careful attention. While interactive segmentation methods can reduce the costs of acquiring segmentation labels for training supervised models, they often still necessitate considerable amounts of ground truth data. Moreover, achieving precise segmentation during the refinement phase results in increased interactions. In this work, we propose an interactive medical segmentation method called PixelDiffuser that requires no medical segmentation ground truth data and only a few clicks to obtain high-quality segmentation using a VGG19-based autoencoder. As the name suggests, PixelDiffuser starts with a small area upon the initial click and gradually detects the target segmentation region. Specifically, we segment the image by creating a distortion in the image and repeating it during the process of encoding and decoding the image through an autoencoder. Consequently, PixelDiffuser enables the user to click a part of the organ they wish to segment, allowing the segmented region to expand to nearby areas with pixel values similar to the chosen organ. To evaluate the performance of PixelDiffuser, we employed the dice score, based on the number of clicks, to compare the ground truth image with the inferred segment. For validation of our method's performance, we leveraged the BTCV dataset, containing CT images of various organs, and the CHAOS dataset, which encompasses both CT and MRI images of the liver, kidneys and spleen. Our proposed model is an efficient and effective tool for medical image segmentation, achieving competitive performance compared to previous work in less than five clicks and with very low memory consumption without additional training.

An Efficient Machine Learning Approach for Prediction of Conjunctiva Hyperemia Assessment using Feature Extraction Methods

The human eye is one of the most intricate sense organs. It is crucial to protect your eyes against several eye disorders that can cause vision loss if untreated in order to maintain your ability to see well. Early detection of eye diseases is therefore crucial in order to prevent any unintended consequences and control the diseases continued progression. Conjunctivitis is one such eye condition that is characterized by conjunctival inflammation, resulting in symptoms like hyperemia (redness) due to increased blood flow. With the aid of the best treatments, modern techniques, and early, precise diagnosis by professionals, it can be cured or can be greatly reduced. The proper diagnosis of the underlying cause of visual problems is frequently postponed or never carried out because of shortage of diagnostic experts, which leads to either insufficient or postponed corrective treatment. In order to diagnose and evaluate conjunctivitis, segmentation methods are essential for locating and measuring hyperemic regions. In the present study, segmentation techniques are applied along with feature extraction techniques to provide an effective machine learning framework for the prediction of eye problems. Using the discrete cosine transform (DCT), the segmented regions of interest are converted into feature vectors. These feature vectors are then used to train machine learning classifiers, including random forest and neural networks, which achieve a promising accuracy of 95.92%. This approach enables ophthalmologists to make more objective and accurate assessments, aiding in disease severity evaluation.

A comparison of structural morphometry in children and adults with persistent developmental stuttering.

This cross-sectional study aimed to differentiate earlier occurring neuroanatomical differences that may reflect core deficits in stuttering versus changes associated with a longer duration of stuttering by analysing structural morphometry in a large sample of children and adults who stutter and age-matched controls. Whole-brain T1-weighted structural scans were obtained from 166 individuals who stutter (74 children, 92 adults; ages 3-58) and 191 controls (92 children, 99 adults; ages 3-53) from eight prior studies in our laboratories. Mean size and gyrification measures were extracted using FreeSurfer software for each cortical region of interest. FreeSurfer software was also used to generate subcortical volumes for regions in the automatic subcortical segmentation. For cortical analyses, separate ANOVA analyses of size (surface area, cortical thickness) and gyrification (local gyrification index) measures were conducted to test for a main effect of diagnosis (stuttering, control) and the interaction of diagnosis-group with age-group (children, adults) across cortical regions. Cortical analyses were first conducted across a set of regions that comprise the speech network and then in a second whole-brain analysis. Next, separate ANOVA analyses of volume were conducted across subcortical regions in each hemisphere. False discovery rate corrections were applied for all analyses. Additionally, we tested for correlations between structural morphometry and stuttering severity. Analyses revealed thinner cortex in children who stutter compared with controls in several key speech-planning regions, with significant correlations between cortical thickness and stuttering severity. These differences in cortical size were not present in adults who stutter, who instead showed reduced gyrification in the right inferior frontal gyrus. Findings suggest that early cortical anomalies in key speech planning regions may be associated with stuttering onset. Persistent stuttering into adulthood may result from network-level dysfunction instead of focal differences in cortical morphometry. Adults who stutter may also have a more heterogeneous neural presentation than children who stutter due to their unique lived experiences.

Deep Network-Assisted Quality Inspection of Laser Welding on Power Battery.

Reliable quality control of laser welding on power batteries is an important issue due to random interference in the production process. In this paper, a quality inspection framework based on a two-branch network and conventional image processing is proposed to predict welding quality while outputting corresponding parameter information. The two-branch network consists of a segmentation network and a classification network, which alleviates the problem of large training sample size requirements for deep learning by sharing feature representations among two related tasks. Moreover, coordinate attention is introduced into feature learning modules of the network to effectively capture the subtle features of defective welds. Finally, a post-processing method based on the Hough transform is used to extract the information of the segmented weld region. Extensive experiments demonstrate that the proposed model can achieve a significant classification performance on the dataset collected on an actual production line. This study provides a valuable reference for an intelligent quality inspection system in the power battery manufacturing industry.

Spatial state analysis of ship during berthing and unberthing process based on 3D LiDAR

In smart ports, accurate perception of vessels' state during berthing and unberthing is crucial to ensure the safety of vessels and ports. In this paper, a method for better monitoring the berthing and unberthing state of vessels is proposed by introducing the GIS spatial analysis technique. For the aspect of the motion state of the vessels, the optimal segmentation region of the hull from the point cloud data is extracted with the aid of GIS spatial theory. Subsequently, the key points are obtained from the optimal segmentation region to calculate the key parameters in the process of berthing and unberthing. Besides, the spatial raster analysis technique is employed to construct the missing point cloud data in the segmentation region. For the other aspect of the spatial state of the vessels, the spatial trajectory of vessels is given. According to the information on a spatial trajectory, a comprehensive study of the spatial state of vessels is performed from the perspective of directional features and spatial relationships. In summary, the proposed method not only identifies the vessels’ motion state from a spatial perspective but also accurately calculates key parameters related to berthing and unberthing. Furthermore, it provides intuitive and clear visualization of the spatial relationship features between the vessel and the shore. Based on the experimental data of berthing and unberthing in Dalian Port, China, accurate results are achieved to verify the feasibility of the proposed method, which can provide a new idea for intelligent berthing and unberthing of vessels.

Impact of segmentation algorithms on multisensor LULC classification in a semiarid Mediterranean area

Geographic object-based image analysis methods usually provide better results than pixel-based methods for classifying land use and land cover from high and medium resolution satellite imagery. This study compares the results of Random Forest (RF) and Multilayer Perceptron (MLP) when used to classify the segments obtained on an RGB+NIR Sentinel-2 image using three different segmentation algorithms, Multiresolution (MR), Region Growing (RG), and Mean-Shift (MS). The hyperparameters of these algorithms were optimised minimising the intra-object heterogeneity and maximizing the inter-object heterogeneity, integrating them in an optimization loop. Geometric and two different centrality and dispersion statistics were computed from some Sentinel-1, Sentinel-2 and LiDAR variables over the segments, and used as features to classify the datasets. The highest segment cross-validation accuracies were obtained with RF using MR segments: 0.9048 (k=0.8905), while the highest accuracies calculated with test pixels were obtained with MLP using MR segments: 0.9447 (k=0.9303), both with the mean and standard deviation of the feature set. Although the overall accuracy is quite high, there are problems with some classes in the confusion matrix and, significant misclassification appear when a qualitative analysis of the final maps is performed, indicating that the accuracy metrics may be overestimated and that a qualitative analysis of the results may also be necessary.

A diagnosis model for brain atrophy using deep learning and MRI of type 2 diabetes mellitus.

Type 2 Diabetes Mellitus (T2DM) is linked to cognitive deterioration and anatomical brain abnormalities like cerebral brain atrophy and cerebral diseases. We aim to develop an automatic deep learning-based brain atrophy diagnosis model to detect, segment, classify, and predict the survival rate. Two hundred thirty-five MRI images affected with brain atrophy due to prolonged T2DM were acquired. The dataset was divided into training and testing (80:20%; 188, 47, respectively). Pre-processing is done through a novel convolutional median filter, followed by segmentation of atrophy regions, i.e., the brain shrinkage, white and gray matter is done through the proposed TRAU-Net model (Transfer Residual Attention U-Net), classification with the proposed Multinomial Logistic regression with Attention Swin Transformer (MLAST), and prediction of chronological age is determined through Multivariate CoX Regression model (MCR). The classification of Brain Atrophy (BA) types is determined based on the features extracted from the segmented region. Performance measures like confusion matrix, specificity, sensitivity, accuracy, F1-score, and ROC-AUC curve are used to measure classification model performance, whereas, for the segmentation model, pixel accuracy and dice similarity coefficient are applied. The pixel accuracy and dice coefficient for segmentation were 98.25 and 96.41, respectively. Brain atrophy multi-class classification achieved overall training accuracy is 0.9632 ± 1.325, 0.9677 ± 1.912, 0.9682 ± 1.715, and 0.9521 ± 1.877 for FA, PA, R-MTA, and L-MTA, respectively. The overall AUC-ROC curve for the classification model is 0.9856. The testing and validation accuracy obtained for the proposed model are 0.9379 and 0.9694, respectively. The prediction model's performance is measured using correlation coefficient (r), coefficient determination r2, and Mean Square Error (MSE) and recorded 0.951, 0.904, and 0.5172, respectively. The brain atrophy diagnosis model consists of sub-models to detect, segment, and classify the atrophy regions using novel deep learning and multivariate mathematical models. The proposed model has outperformed the existing models regarding multi-classification and segmentation; therefore, the automated diagnosis model can be deployed in healthcare centers to assist physicians.

Lifting-based variational multiclass segmentation algorithm: Design, convergence analysis, and implementation with applications in medical imaging

We propose, analyze and realize a variational multiclass segmentation scheme that partitions a given image into multiple regions exhibiting specific properties. Our method determines multiple functions that encode the segmentation regions by minimizing an energy functional combining information from different channels. Multichannel image data can be obtained by lifting the image into a higher dimensional feature space using specific multichannel filtering or may already be provided by the imaging modality under consideration, such as an RGB image or multimodal medical data. Experimental results show that the proposed method performs well in various scenarios. In particular, promising results are presented for two medical applications involving classification of brain abscess and tumor growth, respectively. As main theoretical contributions, we prove the existence of global minimizers of the proposed energy functional and show its stability and convergence with respect to noisy inputs. In particular, these results also apply to the special case of binary segmentation, and these results are also novel in this particular situation.

Deep Learned Segmentations of Inflammation for Novel ⁹⁹mTc-maraciclatide Imaging of Rheumatoid Arthritis.

Rheumatoid arthritis (RA) is an autoimmune disease that causes joint pain, stiffness, and erosion. Power Doppler ultrasound and MRI are imaging modalities used in detecting and monitoring the disease, but they have limitations. ⁹⁹mTc-maraciclatide gamma camera imaging is a novel technique that can detect joint inflammation at all sites in a single examination and has been shown to correlate with power Doppler ultrasound. In this work, we investigate if machine learning models can be used to automatically segment regions of normal, low, and highly inflamed tissue from 192 ⁹⁹mTc-maraciclatide scans of the hands and wrists from 48 patients. Two models were trained: a thresholding model that learns lower and upper threshold values and a neural-network-based nnU-Net model that uses a convolutional neural network (CNN). The nnU-Net model showed 0.94 ± 0.01, 0.51 ± 0.14, and 0.76 ± 0.16 modified Dice scores for segmenting the normal, low, and highly inflamed tissue, respectively, when compared to clinical segmented labels. This outperforms the thresholding model, which achieved modified Dice scores of 0.92 ± 0.01, 0.14 ± 0.07, and 0.35 ± 0.21, respectively. This is an important first step in developing artificial intelligence (AI) tools to assist clinicians' workflow in the use of this new radiopharmaceutical.

Diagnostic performance of CT scan–based radiomics for prediction of lymph node metastasis in gastric cancer: a systematic review and meta-analysis

ObjectiveThe purpose of this study was to evaluate the diagnostic performance of computed tomography (CT) scan–based radiomics in prediction of lymph node metastasis (LNM) in gastric cancer (GC) patients.MethodsPubMed, Embase, Web of Science, and Cochrane Library databases were searched for original studies published until 10 November 2022, and the studies satisfying the inclusion criteria were included. Characteristics of included studies and radiomics approach and data for constructing 2 × 2 tables were extracted. The radiomics quality score (RQS) and Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) were utilized for the quality assessment of included studies. Overall sensitivity, specificity, diagnostic odds ratio (DOR), and area under the curve (AUC) were calculated to assess diagnostic accuracy. The subgroup analysis and Spearman’s correlation coefficient was done for exploration of heterogeneity sources.ResultsFifteen studies with 7,010 GC patients were included. We conducted analyses on both radiomics signature and combined (based on signature and clinical features) models. The pooled sensitivity, specificity, DOR, and AUC of radiomics models compared to combined models were 0.75 (95% CI, 0.67–0.82) versus 0.81 (95% CI, 0.75–0.86), 0.80 (95% CI, 0.73–0.86) versus 0.85 (95% CI, 0.79–0.89), 13 (95% CI, 7–23) versus 23 (95% CI, 13–42), and 0.85 (95% CI, 0.81–0.86) versus 0.90 (95% CI, 0.87–0.92), respectively. The meta-analysis indicated a significant heterogeneity among studies. The subgroup analysis revealed that arterial phase CT scan, tumoral and nodal regions of interest (ROIs), automatic segmentation, and two-dimensional (2D) ROI could improve diagnostic accuracy compared to venous phase CT scan, tumoral-only ROI, manual segmentation, and 3D ROI, respectively. Overall, the quality of studies was quite acceptable based on both QUADAS-2 and RQS tools.ConclusionCT scan–based radiomics approach has a promising potential for the prediction of LNM in GC patients preoperatively as a non-invasive diagnostic tool. Methodological heterogeneity is the main limitation of the included studies.Systematic review registrationhttps://www.crd.york.ac.uk/Prospero/display_record.php?RecordID=287676, identifier CRD42022287676.

Segmentation of Ocular Thermogram Using Level-set Algorithm for Analysis of Contralateral Portions in Healthy Eyes

This work aimed to evaluate the level set segmentation algorithm on ocular surface thermograms. In addition, the vascularity functioning between the contralateral portions of two eyes (right and left) was identified using statistical analysis methods. A total of 25 healthy participants with an average age of 35 years (20 men and 5 women) were selected in April 2022. Thermogram images were captured using a FLIR T series thermal camera. Conventional image processing techniques, such as filtering and edge detection, were used to preprocess thermograms. Next, the level set approach was used with the edge-detected pattern as an input to an automated segmented region of interest (ROI). Five metrics, namely Dice Coefficient, Tanimoto Index, Jaccard Index, Volume Similarity, and Structural Similarity, were used to assess the performance of the segmentation technique compared to ground truth, which showed 97.5%, 92.5%, 94.5%, 96.5%, and 96.5% correlation, respectively, between the segmented and the ground truth images with average values for both the eyes. Statistical analysis demonstrated that the contralateral portions of the ocular thermograms were significantly different in terms of vascular distribution between the left and right eyes (p < 0.005) The level set method efficiently segmented the ROI in ocular thermograms with maximum correlation. According to the segmentation’s results, the model showed the dissimilarity between the contralateral parts of the left and right eyes in healthy cases.

Hybrid Approach to Colony-Forming Unit Counting Problem Using Multi-Loss U-Net Reformulation.

Colony-Forming Unit (CFU) counting is a complex problem without a universal solution in biomedical and food safety domains. A multitude of sophisticated heuristics and segmentation-driven approaches have been proposed by researchers. However, U-Net remains the most frequently cited and used deep learning method in these domains. The latter approach provides a segmentation output map and requires an additional counting procedure to calculate unique segmented regions and detect microbial colonies. However, due to pixel-based targets, it tends to generate irrelevant artifacts or errant pixels, leading to inaccurate and mixed post-processing results. In response to these challenges, this paper proposes a novel hybrid counting approach, incorporating a multi-loss U-Net reformulation and a post-processing Petri dish localization algorithm. Firstly, a unique innovation lies in the multi-loss U-Net reformulation. An additional loss term is introduced in the bottleneck U-Net layer, focusing on the delivery of an auxiliary signal that indicates where to look for distinct CFUs. Secondly, the novel localization algorithm automatically incorporates an agar plate and its bezel into the CFU counting techniques. Finally, the proposition is further enhanced by the integration of a fully automated solution, which comprises a specially designed uniform Petri dish illumination system and a counting web application. The latter application directly receives images from the camera, processes them, and sends the segmentation results to the user. This feature provides an opportunity to correct the CFU counts, offering a feedback loop that contributes to the continued development of the deep learning model. Through extensive experimentation, the authors of this paper have found that all probed multi-loss U-Net architectures incorporated into the proposed hybrid approach consistently outperformed their single-loss counterparts, as well as other comparable models such as self-normalized density maps and YOLOv6, by at least 1% to 3% in mean absolute and symmetric mean absolute percentage errors. Further significant improvements were also reported through the means of the novel localization algorithm. This reaffirms the effectiveness of the proposed hybrid solution in addressing contemporary challenges of precise in vitro CFU counting.