High Segmentation Accuracy Research Articles

CGV-Net: Tunnel Lining Crack Segmentation Method Based on Graph Convolution Guided Transformer

Lining cracking is among the most prevalent forms of tunnel distress, posing significant threats to tunnel operations and vehicular safety. The segmentation of tunnel lining cracks is often hindered by the influence of complex environmental factors, which makes relying solely on local feature extraction insufficient for achieving high segmentation accuracy. To address this issue, this study proposes CGV-Net (CNN, GNN, and ViT networks), a novel tunnel crack segmentation network model that integrates convolutional neural networks (CNNs), graph neural networks (GNNs), and Vision Transformers (ViTs). By fostering information exchange among local features, the model enhances comprehension of the global structural patterns of cracks and improves inference capabilities in recognizing intricate crack configurations. This approach effectively addresses the challenge of modeling contextual information in crack feature extraction. Additionally, the Detailed-Macro Feature Fusion (DMFF) module enables multi-scale feature integration by combining detailed and coarse-grained features, mitigating the significant feature loss encountered during the encoding and decoding stages, and further improving segmentation precision. To overcome the limitations of existing public datasets, which often feature a narrow range of crack types and simplistic backgrounds, this study introduces TunnelCrackDB, a dataset encompassing diverse crack types and complex backgrounds.Experimental evaluations on both the public Crack dataset and the newly developed TunnelCrackDB demonstrate the efficacy of CGV-Net. On the Crack dataset, CGV-Net achieves accuracy, recall, and F1 scores of 73.27% and 57.32%, respectively. On TunnelCrackDB, CGV-Net attains accuracy, recall, and F1 scores of 81.15%, 83.54%, and 82.33%, respectively, showcasing its superior performance in challenging segmentation tasks.

FocusU2Net: Pioneering dual attention with gated U-Net for colonoscopic polyp segmentation.

The detection and excision of colorectal polyps, precursors to colorectal cancer (CRC), can improve survival rates by up to 90%. Automated polyp segmentation in colonoscopy images expedites diagnosis and aids in the precise identification of adenomatous polyps, thus mitigating the burden of manual image analysis. This study introduces FocusU2Net, an innovative bi-level nested U-structure integrated with a dual-attention mechanism. The model integrates Focus Gate (FG) modules for spatial and channel-wise attention and Residual U-blocks (RSU) with multi-scale receptive fields for capturing diverse contextual information. Comprehensive evaluations on five benchmark datasets - Kvasir-SEG, CVC-ClinicDB, CVC-ColonDB, ETISLarib, and EndoScene - demonstrate Dice score improvements of 3.14% to 43.59% over state-of-the-art models, with an 85% success rate in cross-dataset validations, significantly surpassing prior competing models with sub-5% success rates. The model combines high segmentation accuracy with computational efficiency, featuring 46.64 million parameters, 78.09 GFLOPs, and 39.02 GMacs, making it suitable for real-time applications. Enhanced with Explainable AI techniques, FocusU2Net provides clear insights into its decision-making process, improving interpretability. This combination of high performance, efficiency, and transparency positions FocusU2Net as a powerful, scalable solution for automated polyp segmentation in clinical practice, advancing medical image analysis and computer-aided diagnosis.

Segmentation of Breast Lesion using Fuzzy Thresholding and Deep Learning

Breast cancer is a major cause of morbidity and mortality in women. In breast cancer screening, Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCE-MRI) has shown promise as a technique, providing enhanced temporal patterns of breast tissues. This study proposes an enhanced segmentation method for identifying breast lesions. The proposed experiments are done using the breast DCE-MRI images of 123 slices from seven patients in The Cancer Image Archive database. The three methods proposed and tested to segment the lesions are: i) Fuzzy C-mean Thresholding (FCMTH) with morphological operations ii) deep learning networks trained with original images iii) deep learning networks trained with three types of preprocessed images: the core breast image, the filtered core breast image, and the fuzzy thresholded image. In this study, the deep learning networks are trained with preprocessed images generated from the FCMTH technique, resulting in higher segmentation accuracy. FCMTH achieves Dice and Jaccard coefficients of 0.8458 and 0.7471, while DeepLabv3+ with MobileNetv2 trained by preprocessed images achieves 0.9468 and 0.8990, respectively. Thus, the combination of deep learning and FCMTH techniques provides the best performance for lesion detection.

The Measurement of Metal Mineral Particle Size Under the Microscope Based on Gaussian Pyramids and Directional Maximum Intercept

With the development of mineral resources, minerals are becoming increasingly difficult to process. In order to utilize these resources more effectively, in-depth research into process mineralogy has become increasingly important in the field of mineralogy, and particle size measurement under the microscope is one of the critical aspects of process mineralogy. At present, the use of scanning electron microscopes and other equipment for measurement is very expensive, and manual measurement has problems such as poor accuracy and low efficiency. In addition, there is a lack of reference materials for the segmentation algorithm of mineral light images. This article proposes a Gaussian pyramid based on bilateral filtering combined with directional maximum intercept to measure mineral particle size under the microscope. In the experiments, different segmentation algorithms were studied, including Gaussian pyramid segmentation based on bilateral filtering, segmentation based on Fuzzy C-Means, and the rapidly developing deep learning segmentation algorithms in recent years. By comparing the segmentation effects of these three algorithms on various mineral thin-section images, the Gaussian pyramid segmentation algorithm based on bilateral filtering was selected as the optimal one. This was then combined with the directional maximum intercept method to measure the particle size of ilmenite and pyrite images. The experimental results show that the segmentation method based on the bilateral filtering Gaussian pyramid technique has higher segmentation accuracy than the other two algorithms, and can accurately measure the particle size of minerals under the microscope. Compared with manual measurement, this method can effectively and accurately measure the microscopic particle size of target minerals, greatly reducing the workload of measurement personnel and reducing the time spent on measurement.

Clinical feasibility of Ethos auto-segmentation for adaptive whole-breast cancer treatment

IntroductionFollowing a preliminary work validating the technological feasibility of an adaptive workflow with Ethos for whole-breast cancer, this study aims to clinically evaluate the automatic segmentation generated by Ethos.Material and methodsTwenty patients initially treated on a TrueBeam accelerator for different breast cancer indications (right/left, lumpectomy/mastectomy) were replanned using the Ethos® emulator. The adaptive workflow was performed using 5 randomly selected extended CBCTs per patient. The contours generated by artificial intelligence (AI) included both breasts, the heart, and the lungs. The target volumes, specifically the tumor bed (CTV_Boost), internal mammary chain (CTV_IMC), and clavicular nodes (CTV_Nodes), were generated through rigid propagation. The CTV_Breast corresponds to the ipsilateral breast, excluding 5mm from the skin. Two radiation oncologists independently repeated the workflow and qualitatively assessed the accuracy of the contours using a scoring system from 3 (contour to be redone) to 0 (no correction needed). Quantitative evaluation was carried out using the Dice Similarity Coefficient (DSC), Hausdorff Distance (HD), surface Dice (sDSC) and the Added Path Length (APL). The interobserver variability (IOV) between the two observers was also assessed and served as a reference. Lastly, the dosimetric impact of contour correction was evaluated. The physician-validated contours were transferred onto the plans automatically generated by Ethos in adaptive mode. The dose prescription was 52.2Gy in 18 fractions for the boost, 42.3Gy for the breast, IMC, and nodes. The CTV/PTV margin was 2mm for all volumes, except for the IMC (5mm). Dose coverage (D98%) was assessed for the CTVs, while specific parameters for organs at risk (OAR) were evaluated: mean dose and V17Gy (relative volume receiving at least 17Gy) for the ipsilateral lung, mean dose and D2cc (dose received by 2cc volume) for the heart, the contralateral lung and breast.ResultsThe qualitative analysis showed that no correction or only minor corrections were needed for 98.6% of AI-generated contours and 86.7% of the target volumes. Regarding the quantitative analysis, Ethos’ contour generation outperformed inter-observer variability for all structures in terms of DSC, HD, sDSC and APL. Target volume coverage was achieved for 97.9%, 96.3%, 94.2% and 68.8% of the breast, IMC, nodes and boost CTVs, respectively. As for OARs, no significant differences in dosimetric parameters were observed.ConclusionThis study shows high accuracy of segmentation performed by Ethos for breast cancer, except for the CTV_Boost. Contouring practices for adaptive sessions were revised following this study to improve outcomes.

UnifiedCut: A Simple and Efficient Neural Model for Thai, Burmese and Khmer Word Segmentation

Word segmentation is a critical task in natural language processing for southeast Asian Abugida languages, including Thai, Burmese, and Khmer. Existing approaches demonstrate that models using fixed-length windowed context inputs can achieve high segmentation accuracy; however, they often rely on low-level character features or language-specific preprocessing. Character-based methods can limit feature learning, while language-specific features add complexity due to specialized preprocessing requirements. This paper introduces UnifiedCut, which is a neural model that leverages multiple n-grams within a windowed multi-head attention mechanism. This design captures segmentation features from local contexts and multi-perspective n-gram inputs, enhancing generalization and recall, particularly for out-of-vocabulary words. Compared to CNN- and RNN-based approaches, UnifiedCut’s multi-head attention enables finer-grained feature extraction and greater parallelism, resulting in a faster, more scalable solution. Comprehensive experiments on public datasets for Thai, Burmese, and Khmer show that UnifiedCutachieves state-of-the-art performance in word segmentation.

Automated Assessment of Pelvic Longitudinal Rotation Using Computer Vision in Canine Hip Dysplasia Screening.

Canine hip dysplasia (CHD) screening relies on accurate positioning in the ventrodorsal hip extended (VDHE) view, as even mild pelvic rotation can affect CHD scoring and impact breeding decisions. This study aimed to assess the association between pelvic rotation and asymmetry in obturator foramina areas (AOFAs) and to develop a computer vision model for automated AOFA measurement. In the first part, 203 radiographs were analyzed to examine the relationship between pelvic rotation, assessed through asymmetry in iliac wing and obturator foramina widths (AOFWs), and AOFAs. A significant association was found between pelvic rotation and AOFA, with AOFW showing a stronger correlation (R2 = 0.92, p < 0.01). AOFW rotation values were categorized into minimal (n = 71), moderate (n = 41), marked (n = 37), and extreme (n = 54) groups, corresponding to mean AOFA ± standard deviation values of 33.28 ± 27.25, 54.73 ± 27.98, 85.85 ± 41.31, and 160.68 ± 64.20 mm2, respectively. ANOVA and post hoc testing confirmed significant differences in AOFA across these groups (p < 0.01). In part two, the dataset was expanded to 312 images to develop the automated AOFA model, with 80% allocated for training, 10% for validation, and 10% for testing. On the 32 test images, the model achieved high segmentation accuracy (Dice score = 0.96; Intersection over Union = 0.93), closely aligning with examiner measurements. Paired t-tests indicated no significant differences between the examiner and model's outputs (p > 0.05), though the Bland-Altman analysis identified occasional discrepancies. The model demonstrated excellent reliability (ICC = 0.99) with a standard error of 17.18 mm2. A threshold of 50.46 mm2 enabled effective differentiation between acceptable and excessive pelvic rotation. With additional training data, further improvements in precision are expected, enhancing the model's clinical utility.

A CAM-UNet for pose determination of supercapacitor cell module assembly based on monocular vision

Purpose This paper aims to solve the problems of low stacking efficiency and long production time in the supercapacitor module assembly process, a stacking system based on monocular vision is proposed, including bracket visual positioning, grasping and stacking, and it is applied in actual production. Design/methodology/approach To enhance the robustness of the workpiece location method and improve the location accuracy, the improved U-Net network and image processing algorithms are used to segment the collected images. In addition, for the extracted feature points, the objective function that can be globally optimized is obtained by parameterizing the rotation matrix to construct a polynomial equation system and, finally, the equation system is solved to obtain the final pose estimation, which could improve the accuracy of workpiece location. Findings The result indicates that the proposed method is successfully performed on the manipulator. Besides, this method can well solve the problem of object reflection on the conveyor belt. The Intersection over Union of the image segmentation of the object is 0.9948, and the Pixel Accuracy is 0.9973, which has a high segmentation accuracy for the image. The error range between the method proposed in this paper and the pose estimation is within 2 mm, and the qualified rate of supercapacitor module stacking products is over 99.8%. Originality/value This paper proposes a method of accurately extracting feature points by integrating an improved U-Net network and image processing and uses the workpiece positioning algorithm of the optimal solution PnP problem algorithm. The calculation results show that the algorithm improves the positioning accuracy of the workpiece, realizes the assembly of stacked supercapacitor modules and is applied in industrial production.

An attention mechanism-based lightweight UNet for musculoskeletal ultrasound image segmentation.

Accurate musculoseletal ultrasound (MSKUS) image segmentation is crucial for diagnosis and treatment planning. Compared with traditional segmentation methods, deploying deep learning segmentation methods that balance segmentation efficiency, accuracy, and model size on edge devices has greateradvantages. This paper aims to design a MSKUS image segmentation method that has fewer parameters, lower computation complexity and higher segmentationaccuracy. In this study, an attention mechanism-based lightweight UNet (AML-UNet) is designed to segment target muscle regions in MSKUS images. To suppress the transmission of redundant feature, Channel Reconstruction and Spatial Attention Module is designed in the encoding path. In addition, considering the inherent characteristic of MSKUS image, Multiscale Aggregation Module is developed to replace the skip connection architecture of U-Net. Deep supervision is also introduced to the decoding path to refine predicted masks gradually. Our method is evaluated on two MSKUS 2D-image segmentation datasets, including 3917 MSKUS and 1534 images respectively. In the experiments, a five-fold cross-validation method is adopted in ablation experiments and comparison experiments. In addition, Wilcoxon Signed-Rank Test and Bonferroni correction are employed to validate the significance level. 0.01 was used as the statistical significance level in ourpaper. AML-UNet yielded a mIoU of 84.17% and 90.14% on two datasets, representing a 3.38% ( ) and 3.48% ( ) over the Unext model. The number of parameters and FLOPs are only 0.21M and 0.96G, which are 1/34 and 1/29 of those in comparison withUNet. Our proposed model achieved superior results with fewer parameters while maintaining segmentation efficiency and accuracy compared to othermethods.

A multi-scene deep learning model for automated segmentation of acute vertebral compression fractures from radiographs: a multicenter cohort study

ObjectiveTo develop a multi-scene model that can automatically segment acute vertebral compression fractures (VCFs) from spine radiographs.MethodsIn this multicenter study, we collected radiographs from five hospitals (Hospitals A–E) between November 2016 and October 2019. The study included participants with acute VCFs, as well as healthy controls. For the development of the Positioning and Focus Network (PFNet), we used a training dataset consisting of 1071 participants from Hospitals A and B. The validation dataset included 458 participants from Hospitals A and B, whereas external test datasets 1–3 included 301 participants from Hospital C, 223 from Hospital D, and 261 from Hospital E, respectively. We evaluated the segmentation performance of the PFNet model and compared it with previously described approaches. Additionally, we used qualitative comparison and gradient-weighted class activation mapping (Grad-CAM) to explain the feature learning and segmentation results of the PFNet model.ResultsThe PFNet model achieved accuracies of 99.93%, 98.53%, 99.21%, and 100% for the segmentation of acute VCFs in the validation dataset and external test datasets 1–3, respectively. The receiver operating characteristic curves comparing the four models across the validation and external test datasets consistently showed that the PFNet model outperformed other approaches, achieving the highest values for all measures. The qualitative comparison and Grad-CAM provided an intuitive view of the interpretability and effectiveness of our PFNet model.ConclusionIn this study, we successfully developed a multi-scene model based on spine radiographs for precise preoperative and intraoperative segmentation of acute VCFs.Critical relevance statementOur PFNet model demonstrated high accuracy in multi-scene segmentation in clinical settings, making it a significant advancement in this field.Key PointsThis study developed the first multi-scene deep learning model capable of segmenting acute VCFs from spine radiographs.The model’s architecture consists of two crucial modules: an attention-guided module and a supervised decoding module.The exceptional generalization and consistently superior performance of our model were validated using multicenter external test datasets.Graphical

A semi-automatic cardiovascular annotation and quantification toolbox utilizing prior knowledge-guided feature learning

Annotation and quantification of cardiovascular in intravascular ultrasound (IVUS) images are of great significance in assisting the clinical diagnosis and treatment of coronary heart disease. However, it is time-consuming and expensive for cardiologists to manually annotate cardiovascular and calculate parameters in IVUS images using different software during the diagnosis. In this paper, we propose an IVUS analysis toolbox (SaCAQ) for semi-automatic cardiovascular annotation and quantification via prior knowledge-guided feature learning. The SaCAQ consists of two main modules: (1) a semi-automatic cardiovascular annotation module and (2) a cardiovascular quantification module. The semi-automatic cardiovascular annotation module can accurately segment the vascular wall by a IVUS images segmentation model (ISM-IVUS). The ISM-IVUS first extracts vascular morphological features and pixel intensity change features for the complex image segmentation by the expert prior knowledge feature extraction mechanism. Secondly, vascular wall is accurately segmented by the multi-scale feature extraction of attention U-Net. Moreover, the segmentation result checking mechanism is applied to verify the automatic segmentation result, which enhances the reliability of SaCAQ. Plaque regions of interest to doctors are detected by the rule-based plaque detection method, which can assist in the quantification task. Finally, image measurement methods and medically recognized calculation formulas are used in the cardiovascular quantification module to obtain clinical parameters, which significantly improves the confidence of the results. The SaCAQ is verified on our own dataset containing 5242 IVUS slice images and 383 sets of parameters. The experiments demonstrate that the SaCAQ achieves high accuracy of vascular wall segmentation: mIoU of 86.16%, Dice coefficient of 93.43%, and Hausdorff distance of 0.2171 mm. Moreover, ISM-IVUS has good segmentation speed and robustness. The calculated parameters show good correlation and p-values. All of these results indicate that the proposed SaCAQ provides an efficient, accurate, and reliable toolbox for clinical diagnosis of cardiovascular disease. The SaCAQ and user manual can be publicly available from https://github.com/zHwiz/SaCAQ.

SLP-Net:An efficient lightweight network for segmentation of skin lesions

Most existing convolutional neural networks achieve high segmentation accuracy while neglecting the high hardware cost. In this paper, we integrate Spiking neural P-type(SNP-type) and depthwise convolution to introduce a multi-channel SNP-type convolution (MCConvSNP). With it, we propose a lightweight segmentation neural network to assist physicians in precisely identifying lesion areas. The proposed network is an asymmetric network with only encoders, and the decoder is complemented by adaptive fusion and skip connections. It is an asymmetric design that reduces a large number of network parameters. Furthermore, the encoder is composed of an MCConvSNP pyramid, which has a small parameter and enables fast multi-scale feature information extraction. Experiments at ISIC2018 dataset challenge demonstrate that the proposed model has the highest Acc and DSC among the state-of-the-art methods. At the same time, generalization experiments on ISIC2016 dataset and PH2 dataset also demonstrate a favorable generalization ability. Finally, we compare the computational complexity as well as the computational speed of the models in experiments, where SLP-Net has the highest overall superiority.

Analysis of Microscopic Remaining Oil Based on the Fluorescence Image and Deep Learning.

Fossil fuels like oil and natural gas continue to be the primary sources of global energy. Enhancing hydrocarbon recovery from exploited reservoirs has been a major scientific concern in the petroleum industry. Following extended exploitation, the reservoir's oil-water dynamics become intricate, thereby complicating petroleum and natural gas extraction. Pore-scale analysis of microscopic remaining oil (micro-remaining oil) offers theoretical underpinning for enhancing production from high-water-cut oil reservoirs. Fluorescence thin-section analysis allows for the direct evaluation of reservoir oil-bearing properties using oil-containing samples, providing insights into the occurrence and distribution patterns of micro-remaining oil without requiring time-consuming core displacement experiments. The high resolution of fluorescence images further establishes this technique as a representative method for studying micro-remaining oil. However, conventional fluorescence image analysis methods are often subjective and labor-intensive. To address this limitation, we trained four deep learning networks-U-Net, ResU-Net, ScSEU-Net, and Unet++-and applied them innovatively to automate fluorescence image segmentation. Evaluation of network performance via statistical metrics and visual observation indicated that all four networks achieved high segmentation accuracy, particularly ResU-Net, which showed robustness against over-segmentation, under-segmentation, and image noise. Finally, leveraging optimal segmentation results, we conducted quantitative analyses of oil saturation, micro-remaining oil patterns, and pore occupancy. The study demonstrated that ternary composite agents substantially decreased the presence of cluster, film, and adsorbed oils byreducing the oil-water mobility ratio and lowering oil-water interfacial tension. Primarily, these agents displaced crude oil from pores larger than 60 micrometers in an equivalent radius, leading to a significant reduction in their content. Nevertheless, substantial quantities of micro-remaining oil are still confined in pores smaller than 50 micrometers in an equivalent radius, emphasizing the need for attention during subsequent development adjustments. Our research has notably improved the efficiency and accuracy of fluorescence image analysis, effectively supporting the enhancement of recovery in high-water-cut oil reservoirs.

Shape-Aware Adversarial Learning for Scribble-Supervised Medical Image Segmentation with a MaskMix Siamese Network: A Case Study of Cardiac MRI Segmentation.

The transition in medical image segmentation from fine-grained to coarse-grained annotation methods, notably scribble annotation, offers a practical and efficient preparation for deep learning applications. However, these methods often compromise segmentation precision and result in irregular contours. This study targets the enhancement of scribble-supervised segmentation to match the accuracy of fine-grained annotation. Capitalizing on the consistency of target shapes across unpaired datasets, this study introduces a shape-aware scribble-supervised learning framework (MaskMixAdv) addressing two critical tasks: (1) Pseudo label generation, where a mixup-based masking strategy enables image-level and feature-level data augmentation to enrich coarse-grained scribbles annotations. A dual-branch siamese network is proposed to generate fine-grained pseudo labels. (2) Pseudo label optimization, where a CNN-based discriminator is proposed to refine pseudo label contours by distinguishing them from external unpaired masks during model fine-tuning. MaskMixAdv works under constrained annotation conditions as a label-efficient learning approach for medical image segmentation. A case study on public cardiac MRI datasets demonstrated that the proposed MaskMixAdv outperformed the state-of-the-art methods and narrowed the performance gap between scribble-supervised and mask-supervised segmentation. This innovation cuts annotation time by at least 95%, with only a minor impact on Dice performance, specifically a 2.6% reduction. The experimental outcomes indicate that employing efficient and cost-effective scribble annotation can achieve high segmentation accuracy, significantly reducing the typical requirement for fine-grained annotations.

Enhancing sustainable Chinese cabbage production: a comparative analysis of multispectral image instance segmentation techniques

Accurate instance segmentation of individual crops is crucial for field management and crop monitoring in smart agriculture. To address the limitations of traditional remote sensing methods in individual crop analysis, this study proposes a novel instance segmentation approach combining UAVs with the YOLOv8-Seg model. The YOLOv8-Seg model supports independent segmentation masks and detection at different scales, utilizing Path Aggregation Feature Pyramid Networks (PAFPN) for multi-scale feature integration and optimizing sample matching through the Task-Aligned Assigner. We collected multispectral data of Chinese cabbage using UAVs and constructed a high-quality dataset via semi-automatic annotation with the Segment Anything Model (SAM). Using mAP as the evaluation metric, we compared YOLO series algorithms with other mainstream instance segmentation methods and analyzed model performance under different spectral band combinations and spatial resolutions. The results show that YOLOv8-Seg achieved 86.3% mAP under the RGB band and maintained high segmentation accuracy at lower spatial resolutions (1.33 ~ 1.14 cm/pixel), successfully extracting key metrics such as cabbage count and average leaf area. These findings highlight the potential of integrating UAV technology with advanced segmentation models for individual crop monitoring, supporting precision agriculture applications.

Art Pattern Design Based on Visual Semantic Segmentation and Evolutionary Algorithm Optimization

This paper aims to address the issue of traditional art patterns not fully leveraging contextual information in image semantic segmentation. It proposes a multi-level feature aggregation method for image semantic segmentation based on visual Transformer and coordinate attention adjustment (CA Adjustment). Initially, the input image is segmented into slices for linear projection, incorporating learnable positional embeddings to derive an encoded input sequence. Subsequently, employing a visual Transformer-based encoder, the image is transformed into patches to capture global context across the network. CA Adjustment is then introduced to adaptively merge fusion features and visual backbone features using weighted adjustments. Lastly, a multi-objective grasshopper optimization algorithm is designed based on a clustering evolution mechanism to further enhance algorithm performance. Extensive experimental results demonstrate that our proposed algorithm effectively models global contextual information for image feature extraction. Compared to existing advanced algorithms, our method achieves high segmentation accuracy in semantic segmentation tasks, consistently exceeding 95%. Ablation experiments further validate the efficacy of our approach, underscoring its broad applicability in high-precision image semantic segmentation.

Deep learning enables the fully automated quantification of high-resolution cardiac CT images: human in loop for model development

Abstract Introduction Cardiac computed tomography (CCT) images provide vast spatial and temporal information. Segmentation of the structure of interest is necessary to provide quantitative information. However, manual segmentation of cardiac substructures is highly time-consuming. Objective This study aims to develop a robust deep-learning segmentation model for cardiac substructure segmentation in 3D CCT images. Method We initially developed a 3D U-Net architecture, incorporating extensive augmentation, for the segmentation of the Left Ventricle Myocardium (LVM), the blood cavities of the Left and Right Ventricles (LV, RV), and the Left and Right Atria (LA, RA). This model utilized CT images from 20 patients who had available manual segmentation. Subsequently, we applied this initial model to datasets from two centers, including different generations of scanners. We then visually inspected the segmentation, selected the highest-quality segmentations, and modify it if necessary, and excluded the low-quality data. We continued fine-tuning the model with the new dataset until we reached a count of 456 patients. As there were multiple images from one patient during this model's development, we generated datasets to develop a new model using the entire new dataset to avoid any data leakage in the reported metrics. The model was developed using data from Center 1 (5-fold cross-validation) and then evaluated on an external test set from Center 2 (126 patients) and also 4D images from Center 1. Various metrics were calculated to assess the model's performance. Results In the internal test set, utilizing 5-fold CV, the model demonstrated high segmentation accuracy, achieving an overall Dice coefficient of 0.98, with individual scores of 0.96 for LVM, 0.98 for LV, 0.98 for RV, 0.99 for RA, and 0.98 for LA. We assessed the automatic quantification capability of the segmentation model from 256 patients, resulting in an R2 value greater than 0.92 for both LV end-diastolic and systolic volumes. In contrast, the model maintained robust performance on the external dataset, achieving an overall Dice coefficient of 0.96, with individual scores of 0.95 for LVM, 0.98 for LV, 0.97 for RV, 0.98 for RA, and 0.94 for LA. Conclusion We provided a large cohort of ground truth segmentations for different cardiac CT substructures using a human-in-the-loop strategy. Subsequently, we developed a deep learning model for highly accurate automated segmentation of these substructures. This model allows for fast and fully automatic segmentation and quantification of 3D and 4D cardiac CT images, from which various quantitative metrics can be extracted from different cardiac substructures.

Oral screening of dental calculus, gingivitis and dental caries through segmentation on intraoral photographic images using deep learning

ObjectiveIntraoral photographic images are instrumental in the early screening and clinical diagnosis of oral diseases. In addition, people have been trying to apply artificial intelligence to these images. The purpose of this study is to investigate and evaluate a deep learning system designed to segment intraoral photographic images for the detection of dental caries, dental calculus, and gingivitis, and to assess the degree of dental calculus based on the overall features of the tooth surface and gingival margin.Material and methodsThis cross-sectional study collected 3,365 oral endoscopic images, randomly distributed in training datasets (2,019 images), validation dataset (673 images), and test dataset (673 images). The training set and verification set images are manually labeled. An oral endoscopic image segmentation method based on Mamba (Oral-Mamba) and an intelligent evaluation model of dental calculus degree were proposed, achieving the segmentation of two types of oral diseases, namely gingivitis and dental caries, as well as the segmentation of dental calculus regions, and the intelligent evaluation of the degree of dental calculus.ResultsOral-Mamba demonstrated high accuracy in segmentation, with accuracy rates for gingivitis, dental caries, and dental calculus at 0.83, 0.83, and 0.81, respectively. In particular, these rates surpassed those of the U-Net model in IoU, accuracy, and recall metrics. Furthermore, Oral-Mamba runs 25% faster than U-Net.The accuracy of degree classification in the intelligent evaluation model of dental calculus degree is 85%.ConclusionThe proposed deep learning system is expected to be used for the detection of two types of oral diseases and dental calculus, and the degree judgment of photographic images from an intraoral camera. This system offers a practical method to assist in the oral screening of dental caries, dental calculus, and gingivitis, providing benefits such as intuitive use, time efficiency, cost-effectiveness, and ease of deployment.

Periapical lesion detection in periapical radiographs using the latest convolutional neural network ConvNeXt and its integrated models

To overcome the limitation of a single classification model’s inability to simultaneously identify multiple lesion targets within periapical radiographs, This study proposes YoCNET (Yolov5 + ConvNeXt), a novel deep learning integrated model. YoCNET leverages the target detection capability of Yolov5 and the image classification capability of ConvNeXt to achieve automatic segmentation of individual teeth and concurrent detection of periapical lesions across multiple teeth. A dataset of 1,305 periapical radiographs was used to train and validate the ConvNeXt and ResNet34 models, with an 8:2 split for training and validation. Deciduous teeth were excluded from the dataset. Furthermore, 717 individual teeth images were extracted from 200 previously unused periapical radiographs for integrated model validation. Evaluation metrics included accuracy, precision, sensitivity, F1 score, AUC (Area Under Curve), and a confusion matrix.The YoCNET integrated model demonstrated values of 90.93%, 98.88%, 85.30%, 0.9159, and 0.9757 for accuracy, precision, sensitivity, F1 score, and AUC, respectively. These metrics were superior to those achieved by the YoRNET (Yolov5 + ResNet34) integrated model, which recorded 80.47%, 83.78%, 82.16%, 0.8296, and 0.8822. The integrated model achieved high accuracy and efficiency in automatic teeh segmentation by Yolov5 and in automatically detecting multiple periapical lesions by ConvNeXt. YoCNET exhibited superior overall data performance, making it a more suitable deep learning integrated model for clinical applications.

VSegNet - A Variant SegNet for Improving Segmentation Accuracy in Medical Images with Class Imbalance and Limited Data

Deep learning methods for many medical image segmentation task encounter challenges like smaller datasets and class imbalance. This study proposes a variant SegNet (vSegNet) designed to deliver significantly accurate and reliable segmentation results on such datasets. The novelty lies in designing encoder and decoder blocks with an appropriate number of convolution layers and using the Dice score and Hausdorff distance (HD) as compound loss function in learning. This study used public datasets consisting of chest X-rays, axial CT slices, foot ulcer images, and subset of SPIDER dataset to benchmark segmentation task of the proposed neural network model with other popular networks like U-Net, SegNet, DeepLabv3+, VGG16, MobileNetV2, and fully convolutional network (FCN). For the segmentation of lungs in chest X-rays, vertebral body in CT, augmented data for the previous case, foot ulcer dataset, and segmentation of vertebrae, intervertebral disks and spinal canal in SPIDER dataset (MRI dataset) respectively, the proposed vSegNet performed with a Dice score of 0.96 ± 0.01, 0.90 ± 0.20, 0.95 ± 0.02, 0.86 ± 0.07, and 0.95 ± 0.01 and the HD of 14.33 ± 7.74, 8.45 ± 7.08, 7.99 ± 6.05, 29.32 ± 25.64, and 8.45 ± 2.81 with respect to the ground truth on the test dataset. These results highlight the effectiveness of the proposed model in delivering both higher segmentation accuracy and improved boundary delineation. The proposed network, vSegNet has demonstrated as an effective model for semantic segmentation on class-imbalanced smaller datasets, surpassing all other networks considered in this study in terms of mIoU, BF score, Dice score, HD, accuracy, precision, recall and F1 score on a variety of anatomical regions and medical imaging modalities.