Low-contrast Images Research Articles

An Efficient Retinal Fluid Segmentation Network Based on Large Receptive Field Context Capture for Optical Coherence Tomography Images

Optical Coherence Tomography (OCT) is a crucial imaging modality for diagnosing and monitoring retinal diseases. However, the accurate segmentation of fluid regions and lesions remains challenging due to noise, low contrast, and blurred edges in OCT images. Although feature modeling with wide or global receptive fields offers a feasible solution, it typically leads to significant computational overhead. To address these challenges, we propose LKMU-Lite, a lightweight U-shaped segmentation method tailored for retinal fluid segmentation. LKMU-Lite integrates a Decoupled Large Kernel Attention (DLKA) module that captures both local patterns and long-range dependencies, thereby enhancing feature representation. Additionally, it incorporates a Multi-scale Group Perception (MSGP) module that employs Dilated Convolutions with varying receptive field scales to effectively predict lesions of different shapes and sizes. Furthermore, a novel Aggregating-Shift decoder is proposed, reducing model complexity while preserving feature integrity. With only 1.02 million parameters and a computational complexity of 3.82 G FLOPs, LKMU-Lite achieves state-of-the-art performance across multiple metrics on the ICF and RETOUCH datasets, demonstrating both its efficiency and generalizability compared to existing methods.

DeBo: Contrast enhancement for image registration using binary differential evolution and bat optimization.

Image registration has demonstrated its significance as an essential tool for target recognition, classification, tracking, and damage assessment during natural catastrophes. The image registration process relies on the identification of numerous reliable features; thus, low resolutions, poor lighting conditions, and low image contrast substantially diminish the number of dependable features available for registration. Contrast stretching enhances image quality, facilitating the object detection process. In this study, we proposed a hybrid binary differential evolution and BAT optimization model to enhance contrast stretching by optimizing a decision variables in the transformation function. To validate its efficiency, the proposed approach is utilized as a preprocessor before feature extraction in image registration. Cross-comparison of detected features of enhanced images verses the original images during image registration validate the improvements in the image registration process.

Automating an Encoder–Decoder Incorporated Ensemble Model: Semantic Segmentation Workflow on Low-Contrast Underwater Images

Numerous methods have been proposed for semantic segmentation and the state-of-the-art part is likely to be incorporated by deep learning-based methods which show a salient performance. This study addresses the challenge of semantic segmentation in low-contrast imbalanced underwater images. Moreover, it employs nine model fusions as a downstream workflow task using encoder–decoder architectures with Dice Loss and Focal Loss training focusing on the imbalance data. Afterwards, the most effective two encoder–decoder fusion models, Res34+Unet and VGG19+FPN, by 0.592%, 0.590% mIoU on average and by 0.510%, 0.491% F1-score yielded better performance, respectively, than other models. Using a weight-optimization algorithm, the ensemble model with recreated IoU results improves the accuracy for both the Res34+Unet and the VGG19+FPN models, by 0.652% mIoU on average which is 6%. The ensemble model combines the model performances of independent models by considering their superior inference accuracy on a per-class basis separately and improves the model performances by emphasizing the better one on a per-class basis.

A novel network with enhanced edge information for left atrium segmentation from LGE-MRI

IntroductionAutomatic segmentation of the left atrium (LA) constitutes a crucial pre-processing step in evaluating heart structure and function during clinical interventions, such as image-guided radiofrequency ablation of atrial fibrillation. Despite prior research on LA segmentation, the low contrast in medical images exacerbates the challenge of distinguishing various tissues, rendering accurate boundary delineation of the target area formidable. Moreover, class imbalance due to the small target size further complicates segmentation.MethodsThis study aims to devise an architecture that augments edge information for LA segmentation from late gadolinium enhancement magnetic resonance imaging. To intensify edge information within image features, this study introduces an Edge Information Enhancement Module (EIEM) to the foundational network. The design of EIEM is grounded in exploring edge details within target region features learned from images. Additionally, it incorporates a Spatially Weighted Cross-Entropy loss function tailored for EIEM, introducing constraints on different regions based on the importance of pixels to edge segmentation, while also mitigating class imbalance through weighted treatment of positive and negative samples.ResultsThe proposed method is validated on the 2018 Atrial Segmentation Challenge dataset. Compared with other state-of-the-art algorithms, the proposed algorithm demonstrated a significant improvement with an average symmetric surface distance of 0.684 mm and achieved a commendable Dice coefficient of 0.924, implicating the effectiveness of enhancing edge information.DiscussionThe method offers a practical framework for precise LA localization and segmentation, particularly strengthening the algorithm’s effectiveness in improving segmentation outcomes for irregular protrusions and discrete multiple targets. Additionally, the generalizability of our model was evaluated on the heart dataset from the Medical Segmentation Decathlon (MSD) challenge, confirming its robustness across different clinical scenarios involving LA segmentation.

Semi-supervised learning network for deep-sea nodule mineral image segmentation

The accurate segmentation of deep-sea nodule mineral images is crucial for scientific mining. However, due to the low contrast of deep-sea images and the varying sizes of nodule minerals, existing methods are not effective in segmenting these images. Furthermore, fully supervised deep learning methods require numerous labelled images for training, and labelling deep-sea nodule mineral images is highly difficult, resulting in a scarcity of available labeled images, which limits the model generalization ability. To address these challenges, a semi-supervised learning network for deep-sea nodule image segmentation (NmiNet) was proposed. In this method, a semi-supervised training paradigm based on underwater image enhancement perturbation and uncertainty weighted optimization (UEUO) was designed. This paradigm enabled the model to fully mine the features in many unlabeled nodule mineral images under the condition of few labelled nodule mineral images, improving the model generalization ability. Moreover, a lightweight global and local feature extraction (GLFE) module was designed to enhance the attention of the module to small nodules, and its ability to locate nodules of different scales by fusing local and global features without considerably increasing model complexity. Experimental results on deep-sea nodule mineral images reveal that this method outperforms existing approaches.

Accurate Airway Tree Segmentation in CT Scans via Anatomy-aware Multi-class Segmentation and Topology-guided Iterative Learning.

Intrathoracic airway segmentation in computed tomography is a prerequisite for various respiratory disease analyses such as chronic obstructive pulmonary disease, asthma and lung cancer. Due to the low imaging contrast and noises execrated at peripheral branches, the topological-complexity and the intra-class imbalance of airway tree, it remains challenging for deep learning-based methods to segment the complete airway tree (on extracting deeper branches). Unlike other organs with simpler shapes or topology, the airway's complex tree structure imposes an unbearable burden to generate the "ground truth" label (up to 7 or 3 hours of manual or semi-automatic annotation per case). Most of the existing airway datasets are incompletely labeled/annotated, thus limiting the completeness of computer-segmented airway. In this paper, we propose a new anatomy-aware multi-class airway segmentation method enhanced by topology-guided iterative self-learning. Based on the natural airway anatomy, we formulate a simple yet highly effective anatomy-aware multi-class segmentation task to intuitively handle the severe intra-class imbalance of the airway. To solve the incomplete labeling issue, we propose a tailored iterative self-learning scheme to segment toward the complete airway tree. For generating pseudo-labels to achieve higher sensitivity (while retaining similar specificity), we introduce a novel breakage attention map and design a topology-guided pseudo-label refinement method by iteratively connecting breaking branches commonly existed from initial pseudo-labels. Extensive experiments have been conducted on four datasets including two public challenges. The proposed method achieves the top performance in both EXACT'09 challenge using average score and ATM'22 challenge on weighted average score. In a public BAS dataset and a private lung cancer dataset, our method significantly improves previous leading approaches by extracting at least (absolute) 6.1% more detected tree length and 5.2% more tree branches, while maintaining comparable precision.

Radiographic Inspection of Carbon Fiber-Reinforced Polymer Composites (Laminates) with Epoxy and PEEK Binders After Impact and Subsequent Compression Loading.

An approach to detecting discontinuities in carbon fiber-reinforced polymers, caused by impact loading followed by compression testing, was developed. An X-ray sensor-based installation was used, while some algorithms were developed to improve the quality of the obtained low-contrast radiographic images with negligible signal-to-noise ratios. For epoxy/AF (#1) composite subjected to a "high-velocity" steel-ball impact with subsequent compression loading, it was not possible to detect discontinuities since the orientation of the extended zone of interlayer delamination was perpendicular to the irradiation axis. After drop-weight impacts with subsequent compression loading of epoxy/CF (#2) and PEEK/CF (#3) composites, the main cracks were formed in their central parts. This area was reliably detected through the improved radiographic images being more contrasted compared to that for composite #3, for which the damaged area was similar in shape but smaller. The phase variation and congruency methods were employed to highlight low-contrast objects in the radiographic images. The phase variation procedure showed higher efficiency in detecting small objects, while phase congruency is preferable for highlighting large objects. To assess the degree of image improvement, several metrics were implemented. In the analysis of the model images, the most indicative was the PSNR parameter (with a S-N ratio greater than the unit), confirming an increase in image contrast and a decrease in noise level. The NIQE and PIQE parameters enabled the correct assessment of image quality even with the S-N ratio being less than a unit.

Effective low-exposure color image enhancement based on histogram equalization with spatial contextual information

Abstract Color image enhancement refers to improving the visual quality of color images by adjusting various attributes such as brightness, contrast, sharpness, and color balance. This enhancement aims to make the images more visually appealing, precise, and easier to interpret, often by accentuating essential features or details while minimizing noise or distortion. Techniques vary from simple adjustments of brightness and contrast to sophisticated algorithms. Effective image enhancement is crucial in numerous fields like medical imaging; enhancing the medical image is vital for accurate diagnosis and treatment planning, as well as remote sensing for improved satellite and aerial imagery. In security and surveillance, enhanced clarity in footage from surveillance cameras, especially in low-light scenarios, is crucial for identifying subjects and activities. However, image enhancement faces several challenges, like noise amplification and Over-enhancement, leading to unnatural-looking images with exaggerated or distorted features. In this paper, low-exposed or night color images are considered for enhancement, and this paper introduces exposure-based recursive histogram equalization techniques along with an Energy Curve instead of the conventional histogram; the energy curve is similar to a histogram based on spatial Contextual Information of an image. The proposed methods are Recursive Exposure-based Sub-image Histogram Equalization, Recursively Separated Exposure-based Sub-image Histogram Equalization, and Exposure-based Sub-image Histogram Equalization techniques considering spatial contextual information of images using an Energy Curve to improve results. These methods were tested on several publicly available datasetswith low-contrast color images and compared with HE, BBHE, DSHE, CLAHE, ESIHE, R-ESIHE, and RS-ESIHE. The proposed techniques are evaluated using parameterslikeAMBE, PSNR, MSE, Entropy, SSIM, and FSIM. The average PSNR values for eight images (taken from TeV DICeneter https://tev.fbk.eu/resources/imageenhancement) using the techniques mentioned above were 9.10817, 20.80568, 12.82645, 10.39347, 16.27458, 15.19979, and 14.59595, respectively. In contrast, the proposed ESIHE with Energy Curve achieved a PSNR of 22.31585, outperforming other methods across multiple metrics, particularly excelling in noise reduction (PSNR), error minimization (MSE: 193.44), structural similarity (SSIM: 0.733319), and feature retention (FSIM: 0.917678). The comparison demonstrates that enhancement methods can be significantly improved by considering spatial contextual information.

Prior information guided deep-learning model for tumor bed segmentation in breast cancer radiotherapy.

Tumor bed (TB) is the residual cavity of resected tumor after surgery. Delineating TB from CT is crucial in generating clinical target volume for radiotherapy. Due to multiple surgical effects and low image contrast, segmenting TB from soft tissue is challenging. In clinical practice, titanium clips were used as marks to guide the searching of TB. However, this information is limited and may cause large error. To provide more prior location information, the tumor regions on both pre-operative and post-operative CTs are both used by the deep learning model in segmenting TB from surrounding tissues. For breast cancer patient after surgery and going to be treated by radiotherapy, it is important to delineate the target volume for treatment planning. In clinical practice, the target volume is usually generated from TB by adding a certain margin. Therefore, it is crucial to identify TB from soft tissue. To facilitate this process, a deep learning model is developed to segment TB from CT with the guidance of prior tumor location. Initially, the tumor contour on the pre-operative CT is delineated by physician for surgical planning purpose. Then this contour is transformed to the post-operative CT via the deformable image registration between paired pre-operative and post-operative CTs. The original and transformed tumor regions are both used as inputs for predicting the possible region of TB by the deep-learning model. Compared to the one without prior tumor contour information, the dice similarity coefficient of the deep-learning model with the prior tumor contour information is improved significantly (0.812 vs. 0.520, P = 0.001). Compared to the traditional gray-level thresholding method, the dice similarity coefficient of the deep-learning model with the prior tumor contour information is improved significantly (0.812 vs.0.633, P = 0.0005). The prior tumor contours on both pre-operative and post-operative CTs provide valuable information in searching for the precise location of TB on post-operative CT. The proposed method provided a feasible way to assist auto-segmentation of TB in treatment planning of radiotherapy after breast-conserving surgery.

Markerless tracking of tumor and tissues: A motion model approach.

Respiratory motion management is essential in order to achieve high-precision radiotherapy. Markerless motion tracking of tumor can provide a non-invasive way to manage respiratory motion, thereby enhancing treatment accuracy. However, the low contrast in real-time x-ray images for image guidance limits the application of markerless tracking. We present a novel approach based on a motion model to perform markerless tracking of tumor and surrounding tissues even when they have low contrast in real-time x-ray images. A proof-of-concept validation of the method has been performed using digital and physical phantoms at breathing conditions that are significantly different than the planning stage. A motion model is first constructed by performing principal component analysis (PCA) on the planning 4DCT. During treatment, the motion of a surrogate is tracked and used as the input of the motion model, which generates a 3D real-time volume estimation. Such 3D estimation is then projected to 2D to create digitally reconstructed radiographs (DRRs). The relationships between the real-time DRRs, reference DRRs, and reference x-ray images are first established to simulate 2D real-time images from the real-time volume. The registration between the simulated 2D real-time images and real-time x-ray images corrects the initial motion model estimation to ensure the estimated volume matches the real-time condition. In digital phantom, the Dice index of pancreas was improved from 0.74 to 0.78 after correction using real-time DRRs in fully inhaled phase. Validation on lung and pancreas is performed in physical phantom with two motion traces. The surrogate-tumor relationships were intentionally altered to generate large target localization errors due to the differences in body condition between treatment planning stage and during treatment. The real-time correction for the estimated 3D real-time volume was performed using a pair of 2D x-ray images. For the deep breathing motion trace, the tumor localization mean absolute error (MAE) throughout the tracking decreases from around 3mm to less than 1mm after correction. For the shallow breathing motion trace with a 1.7mm baseline shift, the tumor localization MAE throughout the tracking decreases from around 1.5mm to less than 1mm after correction. The method combines the detailed structural information from planning 4DCT and real-time information from real-time x-ray images through a motion model. The matching between the real-time model estimation and 2D real-time images is performed in the same modality so that it can be applied to regions with low contrast in the images. The real-time images successfully corrected the initial motion model estimations in our proof-of-concept validation. This suggests the potential to perform markerless tracking in low-contrast region using a motion model.

Breast Lesion Detection for Ultrasound Images Using MaskFormer.

This study evaluates the performance of the MaskFormer model for segmenting and classifying breast lesions using ultrasound images, addressing ultrasound's limitations. Ultrasound used for breast cancer detection faces challenges like low image contrast and difficulty in the detection of small or multiple lesions, further complicated by variability based on operator skill. Initial experiments with U-Net and other CNN-based models revealed constraints, such as early plateauing in model loss, which indicated suboptimal learning and performance. In contrast, MaskFormer demonstrated continuous improvement, achieving higher precision in breast lesion segmentation and significantly reducing both false positives and false negatives. Comparative analysis showed MaskFormer's superior performance, with the highest precision and recall rates for malignant lesions and an overall mean average precision (mAP) of 0.943. The model's ability to detect a diverse range of breast lesions, including those potentially missed by the human eye, especially by less experienced practitioners, underscores its potential. These findings suggest that integrating AI models like MaskFormer could greatly enhance ultrasound performance for breast cancer detection, providing reliable, operator-independent image analysis and potentially improving patient outcomes on a global scale.

X-ray coronary angiography background subtraction by adaptive weighted total variation regularized online RPCA

Objective.X-ray coronary angiograms (XCA) are widely used in diagnosing and treating cardiovascular diseases. Various structures with independent motion patterns in the background of XCA images and limitations in the dose of injected contrast agent have resulted in low-contrast XCA images. Background subtraction methods have been developed to enhance the visibility and contrast of coronary vessels in XCA sequences, consequently reducing the requirement for excessive contrast agent injections.Approach.The current study proposes an adaptive weighted total variation regularized online RPCA (WTV-ORPCA) method, which is a low-rank and sparse subspaces decomposition approach to subtract the background of XCA sequences. In the proposed method, the images undergo initial preprocessing using morphological operators to eliminate large-scale background structures and achieve image homogenization. Subsequently, the decomposition algorithm decomposes the preprocessed images into background and foreground subspaces. This step applies an adaptive weighted TV constraint to the foreground subspace to ensure the spatial coherency of the finally extracted coronary vessel images.Main results.To evaluate the effectiveness of the proposed background subtraction method, some qualitative and quantitative experiments are conducted on two clinical and synthetic low-contrast XCA datasets containing videos from 21 patients. The obtained results are compared with six state-of-the-art methods employing three different assessment criteria. By applying the proposed method to the clinical dataset, the mean values of the global contrast-to-noise ratio, local contrast-to-noise ratio, structural similarity index, and reconstruction error (RE) are obtained as5.976,3.173,0.987, and0.026, respectively. These criteria over the low-contrast synthetic dataset were4.851,2.942,0.958, and0.034, respectively.Significance.The findings demonstrate the superiority of the proposed method in improving the contrast and visibility of coronary vessels, preserving the integrity of the vessel structure, and minimizing REs without imposing excessive computational complexity.

Illumination Correction and Enhancement Algorithm for Defective Powder Spreading Images in Selective Laser Sintering

Abstract Based on the uneven illumination distribution and low defect contrast in the powder spreading images acquired during the selective laser sintering manufacturing process, which affects subsequent defect detection, an adaptive nonlinear illumination correction and enhancement algorithm is proposed. Local illumination correction is performed using gray valueitting curves and is combined with global gamma correction to improve correction capability in overly dark areas. A nonlinear enhancement function is constructed to boost the contrast of the image’s illumination component. Finally, a gradient-based L0 norm minimization algorithm is used to smooth non-defect areas. Experimental results show that the proposed algorithm signi icantly improves correction and enhancement, outperforming comparison algorithms overall. Compared to the original images, the coef icient of variation decreases by an average of 0.179, and the standard deviation increases by an average of 26.7%, demonstrating the effectiveness of the proposed algorithm.

Infrared Weak Target Detection in Dual Images and Dual Areas

This study proposes a novel approach for detecting weak small infrared (IR) targets, called double-image and double-local contrast measurement (DDLCM), designed to overcome challenges of low contrast and complex backgrounds in images. In this approach, the original image is decomposed into odd and even images, and the gray difference contrast is determined using a dual-neighborhood sliding window structure, enhancing target saliency and contrast by increasing the distinction between the target and the local background. A central unit is then constructed to capture relationships between neighboring and non-neighboring units, aiding in clutter suppression and eliminating bright non-target interference. Lastly, the output value is derived by extracting the lowest contrast value of the weak small targets from the saliency map in each direction. Experimental results on two datasets demonstrate that the DDLCM algorithm significantly enhances real-time IR dim target detection, achieving an average performance improvement of 32.83%. The area under the ROC curve (AUC) decline is effectively controlled, with a maximum reduction limited to 3%. Certain algorithms demonstrate a notable AUC improvement of up to 43.96%. To advance infrared dim target detection research, we introduce the IFWS dataset for benchmarking and validating algorithm performance.

DAAM-Net: A dual-encoder U-Net network with adjacent auxiliary module for pituitary tumor and jaw cyst segmentation

In the diagnosis of various diseases, accurate segmentation of lesions in medical images is crucial. However, when faced with the challenges of blurred edges, noise, and low contrast in images of pituitary tumors and jaw cysts, existing methods do not perform well in solving these problems. To overcome these problems, we present a new approach based on U-Net architecture. Firstly, a more comprehensive semantic information capture is realized by superimposing two U-Net network structures. Following that, the method incorporates atrous spatial pyramid pooling (ASSP) technology to enable adjustable multi-scale feature extraction. Thirdly, an adjacent auxiliary module (AAM) is designed to smoothly fuse features from different levels, effectively suppressing noise interference and reducing feature ambiguity. By testing on the dataset provided by Quzhou People’s Hospital, the IoU, Dice, Specificity, and Mcc of the method were 77.46%, 87.18%, 99.64% and 86.97% on the pituitary tumor dataset, and 81.32%, 89.53%, 99.91% and 89.55% on the jaw cyst dataset, respectively. Compared with existing algorithms, our method exhibits superior segmentation performance and is expected to become a more reliable auxiliary tool in the diagnosis of medical diseases.

CRML-Net: Cross-Modal Reasoning and Multi-Task Learning Network for tooth image segmentation

Data from a single modality may suffer from noise, low contrast, or other imaging limitations that affect the model’s accuracy. Furthermore, due to the limited amount of data, most models trained on single-modality data tend to overfit the training set and perform poorly on out-of-domain data. Therefore, in this paper, we propose a network named Cross-Modal Reasoning and Multi-Task Learning Network (CRML-Net), which combines cross-modal reasoning and multi-task learning, aiming to leverage the complementary information between different modalities and tasks to enhance the model’s generalization ability and accuracy. Specifically, CRML-Net consists of two stages. In the first stage, our network extracts a new morphological information modality from the original image and then performs cross-modal fusion with the original modality image, aiming to leverage the morphological information to enhance the model’s robustness to out-of-domain datasets. In the second stage, based on the output of the previous stage, we introduce a multi-task learning mechanism, aiming to improve the model’s performance on unseen data by sharing surface detail information from auxiliary tasks. We validated our method on a publicly available tooth cone beam computed tomography dataset. Our evaluation demonstrates that our method outperforms state-of-the-art approaches.

Thermal Battery Multi-Defects Detection and Discharge Performance Analysis Based on Computed Tomography Imaging

To address the typical structural defects that are prone to occur during the preparation and storage processes of thermal battery, experiments of battery image acquisition were designed based on X-ray computed tomography system. An improved Yolov5s network was employed to achieve high-precision automatic detection of typical defects. Through the discharge experiment of thermal battery, discharge performance curves of normal batteries and three defective batteries were constructed. The impact and mechanisms of different defects on the discharge performance were analyzed based on the voltage curve. By designing an automatic stitching scheme, the phenomenon of interlayer information overlap caused by the increase of cone angle in digital radiography images was suppressed. To address the issues of low image contrast and limited defect data in thermal battery imaging, the defect dataset was expanded using the designed image preprocessing steps and improving the contrast of the images. For subtle defects that are difficult to identify, the introduced multi-head self-attention mechanism in Transformer and the use of Focal Loss instead of cross-entropy loss function were employed to improve the recognition accuracy of subtle defects while ensuring the detection speed. The comparative experiment shows that the improved network model has higher recognition accuracy compared to Faster R-CNN, SSD, Cascade R-CNN, EfficientDet and the original Yolov5s network. The recognition accuracy of typical defects in thermal batteries can reach 98.7%.

Enhancing Image Contrast in Breast USCT Reflection Imaging Using Coherence Factor Beamforming

Abstract Breast cancer is the most common cancer worldwide, and early screening is essential. Existing breast ultrasound is affordable and convenient, but its 2D image quality is poor, and it heavily relies on the doctor’s skills. The novel breast ultrasound computed tomography (USCT) can use a ring array probe to implement hundreds of single transducer emission for full-view focusing synthetic aperture reflection imaging, thereby obtaining high-resolution 3D images. However, single transducer emission results in low acoustic intensity, which further leads to low signal-to-noise ratio and low contrast of reflection images. To improve image contrast, this study incorporated beamforming with coherence factor (CF) into reflection imaging using ring transducers, promising to produce high-contrast USCT reflection images. For evaluation, we developed a USCT prototype with 2048 elements and conducted phantom experiments. We conducted phantom experiments. Results showed that the contrast achieved by the coherence factor method of the breast phantom image is improved from 24.54 dB to 63.18 dB, and the artifacts are significantly suppressed. It can be seen that the CF method could significantly improve the contrast of the breast USCT reflection image.

CafeNet: A Novel Multi‐Scale Context Aggregation and Multi‐Level Foreground Enhancement Network for Polyp Segmentation

ABSTRACTThe detection of polyps plays a significant role in colonoscopy examinations, cancer diagnosis, and early patient treatment. However, due to the diversity in the size, color, and shape of polyps, as well as the presence of low image contrast with the surrounding mucosa and fuzzy boundaries, precise polyp segmentation remains a challenging task. Furthermore, this task requires excellent real‐time performance to promptly and efficiently present predictive results to doctors during colonoscopy examinations. To address these challenges, a novel neural network, called CafeNet, is proposed in this paper for rapid and accurate polyp segmentation. CafeNet utilizes newly designed multi‐scale context aggregation (MCA) modules to adapt to the extensive variations in polyp morphology, covering small to large polyps by fusing simplified global contextual information and local information at different scales. Additionally, the proposed network utilizes newly designed multi‐level foreground enhancement (MFE) modules to compute and extract differential features between adjacent layers and uses the prediction output from the adjacent lower‐layer decoder as a guidance map to enhance the polyp information extracted by the upper‐layer encoder, thereby improving the contrast between polyps and the background. The polyp segmentation performance of the proposed CafeNet network is evaluated on five benchmark public datasets using six evaluation metrics. Experimental results indicate that CafeNet outperforms the state‐of‐the‐art networks, while also exhibiting the least parameter count along with excellent real‐time operational speed.

STU3Net: An Improved U‐Net With Swin Transformer Fusion for Thyroid Nodule Segmentation

ABSTRACTThyroid nodules are a common endocrine system disorder for which accurate ultrasound image segmentation is important for evaluation and diagnosis, as well as a critical step in computer‐aided diagnostic systems. However, the accuracy and consistency of segmentation remains a challenging task due to the presence of scattering noise, low contrast and resolution in ultrasound images. Therefore, we propose a deep learning‐based CAD (computer‐aided diagnosis) method, STU3Net in this paper, aiming at automatic segmentation of thyroid nodules. The method employs a modified Swin Transformer combined with a CNN encoder, which is capable of extracting morphological features and edge details of thyroid nodules in ultrasound images. In decoding through the features for image reconstruction, we introduce a modified three‐layer U‐Net network with cross‐layer connectivity to further enhance image reduction. This cross‐layer connectivity enhances the network's capture and representation of the contained image feature information by creating skip connections between different layers and merging the detailed information of the shallow network with the abstract information of the deeper network. Through comparison experiments with current mainstream deep learning methods on the TN3K and BUSI datasets, we validate the superiority of the STU3Net method in thyroid nodule segmentation performance. The experimental results show that STU3Net outperforms most of the mainstream models on the TN3K dataset, with Dice and IoU reaching 0.8368 and 0.7416, respectively, which are significantly better than other methods. The method demonstrates excellent performance on these datasets and provides radiologists with an effective auxiliary tool to accurately detect thyroid nodules in ultrasound images.