Image Size Research Articles

A novel compression methodology for medical images using deep learning for high-speed transmission

Medical imaging is a rapidly growing field having a high impact on the early detection, diagnosis and surgical planning of diseases. Several imaging techniques such as computed tomography (CT), magnetic resonance imaging (MRI) and ultrasound (US) imaging generate a higher volume of data, necessitating additional storage and communication requirements. Hence, image compression is utilized in medical field to reduce redundancy and alleviate memory and bandwidth issues. This paper presents a novel deep learning-based compression method to reduce the size of medical images. This method employs a deep convolutional neural network for learning compact representations of medical images, then coded by a Huffman encoder. The compression process is reversed to reconstruct the original image. Several tests are conducted to compare the results with other wellknown compression methods. The proposed model achieved a mean peak signal-to-noise ratio (PSNR) of 42.82 dB with storage space saving (SSS) of 96.15% for CT, 43.88 dB with SSS of 96.25% for MRI, 46.29 dB with SSS of 96.07% for US and 43.51 dB with SSS of 96.95% for X-ray images. The findings showed that the proposed compression technique could greatly compress the image size, saving storage space, facilitating better transmission and preserving critical diagnostic information.

CrowdUNet: Segmentation assisted U-shaped crowd counting network

With the end of the COVID-19 pandemic, the number of pedestrians in various public places has increased dramatically. Estimating the size and density distribution of crowds accurately from images is essential for public safety. At present, there are still many factors that limit the accuracy of dense crowd counting, such as perspective distortion, background clutter and heavy occlusion. To be capable of accurately estimating the crowd size in RGB images, we propose a crowd counting network called CrowdUNet, which is assisted by a segmentation task. It applies the segmentation results to the crowd counts, making the network more focused on the prediction of foreground regions. We combine Swin Transformer Block and CNN to build a Swin Transformer Convolution(STC) module to extract deep semantic features. We analyze the characteristics of crowd images and propose a novel decoder structure called Coordinate Decoder(CD), which better aggregates low and high level features and improve the robustness of the network. In order to obtain accurate regression results, we also propose a regression head with multi-scale receptive fields, which is called Spatial Pyramid Convolution (SPC). Extensive experiments on four challenging crowd counting datasets namely ShanghaiTech A, ShanghaiTech B, UCF p=CC 50, and UCF-QNRF have validated the proposed method.

MRCM‐UCTransNet: Automatic and Accurate 3D Tooth Segmentation Network From Cone‐Beam CT Images

ABSTRACTMany scenarios in dental clinical diagnosis and treatment require the segmentation and identification of a specific tooth or the entire dentition in cone‐beam computed tomography (CBCT) images. However, traditional segmentation methods struggle to ensure accuracy. In recent years, there has been significant progress in segmentation algorithms based on deep learning, garnering considerable attention. Inspired by models from present neuro networks such as UCTransNet and DC‐Unet, this study proposes an MRCM‐UCTransNet for accurate three‐dimensional tooth segmentation from cone‐beam CT images. To enhance feature extraction while preserving the multi‐head attention mechanism, a multi‐scale residual convolution module (MRCM) is integrated into the UCTransNet architecture. This modification addresses the limitations of traditional segmentation methods and aims to improve accuracy in tooth segmentation from CBCT images. Comparative experiments indicate that, in the situation with a specific image size and small data volume, the proposed method exhibits certain advantages in segmentation accuracy and precision. Compared to traditional Unet approaches, MRCM‐UCTransNet's dice accuracy is improved by 7%, while its sensitivity is improved by about 10%. These findings highlight the efficacy of the proposed approach, particularly in scenarios with specific image size constraints and limited data availability. The proposed MRCM‐UCTransNet algorithm integrates the latest architectural advancements in the Unet model which achieves effective segmentation of six types of teeth within the tooth. It was proved to be efficient for image segmentation on small datasets, requiring less training time and fewer parameters.

A Survey on Visual Mamba

State space models (SSM) with selection mechanisms and hardware-aware architectures, namely Mamba, have recently shown significant potential in long-sequence modeling. Since the complexity of transformers’ self-attention mechanism is quadratic with image size, as well as increasing computational demands, researchers are currently exploring how to adapt Mamba for computer vision tasks. This paper is the first comprehensive survey that aims to provide an in-depth analysis of Mamba models within the domain of computer vision. It begins by exploring the foundational concepts contributing to Mamba’s success, including the SSM framework, selection mechanisms, and hardware-aware design. Then, we review these vision Mamba models by categorizing them into foundational models and those enhanced with techniques including convolution, recurrence, and attention to improve their sophistication. Furthermore, we investigate the widespread applications of Mamba in vision tasks, which include their use as a backbone in various levels of vision processing. This encompasses general visual tasks, medical visual tasks (e.g., 2D/3D segmentation, classification, image registration, etc.), and remote sensing visual tasks. In particular, we introduce general visual tasks from two levels: high/mid-level vision (e.g., object detection, segmentation, video classification, etc.) and low-level vision (e.g., image super-resolution, image restoration, visual generation, etc.). We hope this endeavor will spark additional interest within the community to address current challenges and further apply Mamba models in computer vision.

Adaptive non-iterative histogram-based hologram quantization

Quantization is widely used for compression, storage, transmission and display of computer-generated and digital holograms. It reduces number of hologram gradations to obtain better ratio of the quality of reconstructed 2D and 3D objects to the size of holographic images or videos. This paper proposes 2 methods of non-iterative quantization of holograms: adaptive non-uniform quantization based on histogram processing with peak shift (ANHS method) and additional application of logarithmic companding (ANHSL method). They were compared with 4 methods of iterative and non-iterative quantization on holograms with sizes up to 2048 × 2048 pixels. The proposed methods showed the best reconstruction quality that is higher up to 59 % in comparison with non-iterative uniform quantization. Especially the proposed methods give higher quality at little number of hologram gradations (from 2 to 16 gradations). In comparison with the most qualitative iterative methods (Lloyd-Max and k-means) the quality is increased up to 9 %, and processing time is reduced in 10–1000 times.

Air Pollution Monitoring Model in Real-time with cloud based Channel Ranging and Stego Privacy Preservation

<p>Monitoring air quality in polluted environments is crucial for human health, but traditional methods using meteorological equipment have limitations in complex terrains and are costly. This study proposes a novel approach to address air pollution monitoring in urban areas by visually assessing the haziness of distant photos. We explore the correlation between air quality indexes (e.g., AQI, PM2.5, PM10) and haziness levels in monitoring images. Results indicate that objective indicators accurately reflect air pollution levels, regardless of image size. To implement this observation practically, we introduce a new method called the "Channel ranging based prediction model in cloud with steganography-based privacy preservation framework." This model calculates a ratio between dark and bright channel information in scaled images, serving as a visual index of air pollution. Experimental findings demonstrate the effectiveness of this metric in terms of correlation, computational speed, and classification accuracy compared to existing metrics.</p>

Optimizing Performance of Transformer-based Models for Fetal Brain MR Image Segmentation.

Purpose To test the performance of a transformer-based model when manipulating pretraining weights, dataset size, and input size and comparing the best model with the reference standard and state-of-the-art models for a resting-state functional (rs-fMRI) fetal brain extraction task. Materials and Methods An internal retrospective dataset (172 fetuses, 519 images; collected 2018-2022) was used to investigate influence of dataset size, pretraining approaches, and image input size on Swin-U-Net transformer (UNETR) and UNETR models. The internal and external (131 fetuses, 561 images) datasets were used to cross-validate and to assess generalization capability of the best model versus state-of-the-art models on different scanner types and number of gestational weeks (GWs). The Dice similarity coefficient (DSC) and the balanced average Hausdorff distance (BAHD) were used as segmentation performance metrics. Generalized equation estimation multifactorial models were used to assess significant model and interaction effects of interest. Results The Swin-UNETR model was not affected by the pretraining approach and dataset size and performed best with the mean dataset image size, with a mean DSC of 0.92 and BAHD of 0.097. Swin-UNETR was not affected by scanner type. Generalization results on the internal dataset showed that Swin-UNETR had lower performance compared with the reference standard models and comparable performance on the external dataset. Cross-validation on internal and external test sets demonstrated better and comparable performance of Swin-UNETR versus convolutional neural network architectures during the late-fetal period (GWs > 25) but lower performance during the midfetal period (GWs ≤ 25). Conclusion Swin-UNTER showed flexibility in dealing with smaller datasets, regardless of pretraining approaches. For fetal brain extraction from rs-fMR images, Swin-UNTER showed comparable performance with that of reference standard models during the late-fetal period and lower performance during the early GW period. Keywords: Transformers, CNN, Medical Imaging Segmentation, MRI, Dataset Size, Input Size, Transfer Learning Supplemental material is available for this article. © RSNA, 2024.

Skin lesion segmentation via Neural Cellular Automata

Skin melanoma is one of the most dangerous tumor lesions. In recent years, the number of cases and deaths caused by melanoma has been increasing. The discovery and segmentation of the lesion area are crucial for the timely diagnosis and treatment of melanoma. However, the lesion area is often similar to the healthy area, and the size scale changes greatly, which makes the segmentation of the skin lesion area a very challenging task. Neural Cellular Automata (NCA) is a model that can be described as a recurrent convolutional network. It achieves global consistency through multiple local information exchanges, thereby completing the processing of information. Recent research on NCA shows that such a local interactive model can segment low-resolution images well. However, for high-resolution images, direct use of NCA for processing is limited by high memory requirements and difficulty in model convergence. In this paper, in order to overcome these limitations, we propose a new NCA-based segmentation model, named UNCA. UNCA is a model with a U-shaped structure. The high-resolution image is down-sampled to obtain high-dimensional low-resolution features, which are then input into the NCA for information processing. Finally, the image size is restored by upsampling. The experimental results show that the UNCA proposed in this paper has achieved good results on the ISIC2017 dataset, surpassing most of the current methods.

Performance measurement implementation on the smart fisheries village website using pagespeed insight

Websites have become the primary way organizations and individuals to communicate, provide information, and offer daily services. The purpose of creating the Smart Fisheries Village (SFV) website was to enhance the performance and quality of the user experience by measuring and optimizing image sizes using Google's tools, specifically Google PageSpeed Insight. We monitored and analyzed the implementation performance to ensure faster loading times without compromising visual quality. The implementation results showed significant improvements in the SFV loading speed, leading to a more satisfactory user experience. To identify images that slow website loading, we used data from PageSpeed Insight. After implementing improvements, we distributed a questionnaire to users to evaluate the development results. The results of the questionnaire revealed a significant increase in user satisfaction with the loading speed and quality of the user experience of the Bangsring Smart Fisheries Village (SFV) website. These findings provide valuable information for the continued development and optimization of website performance in the future. Therefore, this research makes a valuable contribution to improving the performance and user experience of the Bangsring Smart Fisheries Village (SFV) website.

Deep cell phenotyping and spatial analysis of multiplexed imaging with TRACERx-PHLEX

The growing scale and dimensionality of multiplexed imaging require reproducible and comprehensive yet user-friendly computational pipelines. TRACERx-PHLEX performs deep learning-based cell segmentation (deep-imcyto), automated cell-type annotation (TYPEx) and interpretable spatial analysis (Spatial-PHLEX) as three independent but interoperable modules. PHLEX generates single-cell identities, cell densities within tissue compartments, marker positivity calls and spatial metrics such as cellular barrier scores, along with summary graphs and spatial visualisations. PHLEX was developed using imaging mass cytometry (IMC) in the TRACERx study, validated using published Co-detection by indexing (CODEX), IMC and orthogonal data and benchmarked against state-of-the-art approaches. We evaluated its use on different tissue types, tissue fixation conditions, image sizes and antibody panels. As PHLEX is an automated and containerised Nextflow pipeline, manual assessment, programming skills or pathology expertise are not essential. PHLEX offers an end-to-end solution in a growing field of highly multiplexed data and provides clinically relevant insights.

Light field imaging technology for virtual reality content creation: A review

AbstractThe light field (LF) imaging technique can capture 3D scene information in 4D by recording both 2D intensity and 2D direction of incoming light rays. Due to this capability, LF has shown a great interest in virtual reality (VR) and augmented reality (AR) for enhanced immersion, improved depth perception and reconstruction of realistic 3D environments. This paper presents a comprehensive review of LF imaging technology and other approaches used for VR content creation. The applications of LF technology beyond VR and AR are also discussed. The challenges and limitations of other approaches for VR content creation are examined. State‐of‐the‐art research has focused on how VR experiences benefit from LF technology and identified the challenges to creating comfortable, immersive and realistic VR content such as (1) image size and resolution, (2) processing speed, (3) precise calibration and (4) depth reconstruction. Recommendations that can be considered for creating immersive VR content are provided to enhance user experience. These recommendations aim to contribute to developing more comfortable and realistic VR content, extending the potential applications of LF imaging technology in diverse fields.

DiffRank: Enhancing efficiency in discontinuous frame rate analysis for urban surveillance systems

Urban public safety management relies heavily on video surveillance systems, which provide crucial visual data for resolving a wide range of incidents and controlling unlawful activities. Traditional methods for target detection predominantly employ a two-stage approach, focusing on precision in identifying objects such as pedestrians and vehicles. These objects, typically sparse in large-scale, lower-quality surveillance footage, induce considerable redundant computation during the initial processing stage. This redundancy constrains real-time detection capabilities and escalates processing costs. Furthermore, transmitting raw images and videos laden with superfluous information to centralized back-end systems significantly burdens network communications and fails to capitalize on the computational resources available at diverse surveillance nodes. This study introduces DiffRank, a novel preprocessing method for fixed-angle video imagery in urban surveillance. The method strategically generates candidate regions during preprocessing, thereby reducing redundant object detection and improving the efficiency of the detection algorithm. Drawing upon change detection principles, a background feature learning approach utilizing shallow features has been developed. This approach prioritizes learning the characteristics of fixed-area backgrounds over direct background identification. As a result, alterations in ROI are efficiently discerned using computationally efficient shallow features, markedly accelerating the generation of proposed Regions of Interest (ROIs) and diminishing the computational demands for subsequent object detection and classification. Comparative analysis on various public and private datasets illustrates that DiffRank, while maintaining high accuracy, substantially outperforms existing baselines in terms of speed, particularly with larger image sizes (e.g., an improvement exceeding 300 % at 1920×1080 resolution). Moreover, the method demonstrates enhanced robustness compared to baseline methods, efficiently disregarding static targets like mannequins in display windows. The advancements in candidate area preprocessing enable a balanced approach between detection accuracy and overall detection speed, making the algorithm highly applicable for real-time on-site analysis in edge computing scenarios and cloud-edge collaborative computing environments.

Unishyper: A Rust-based unikernel enhancing reliability and efficiency of embedded systems

Unikernels are simple, customizable, efficient, and small in code size, which makes them highly applicable to embedded scenarios. However, most existing unikernels are developed and optimized for cloud computing, and they do not fully meet the requirements of high reliability and platform customization in embedded environments. We propose Unishyper, a reliable and high-performance embedded unikernel in Rust. To support memory isolation between user applications, user code, and kernel code, Unishyper designs the Zone mechanism on top of Intel MPK. Unishyper further proposes a thread-level unwind strategy for safe fault handling while avoiding memory leakage. Finally, Unishyper supports fine-grained customization, seamlessly integrates with the Rust ecosystem, and uses Unilib for function offloading to further reduce image size. Our evaluation results show that Unishyper achieves better performance than peer unikernels on major micro-benchmarks, can effectively stop illegal memory accesses across application boundaries, and has a minimal memory footprint of less than 100 KB.

Higher-contrast images are better remembered during naturalistic encoding

It is unclear whether memory for images of poorer visibility (as low contrast or small size) will be lower due to weak signals elicited in early visual processing stages, or perhaps better since their processing may entail top-down processes (as effort and attention) associated with deeper encoding. We have recently shown that during naturalistic encoding (free viewing without task-related modulations), for image sizes between 3°–24°, bigger images stimulating more visual system processing resources at early processing stages are better remembered. Similar to size, higher contrast leads to higher activity in early visual processing. Therefore, here we hypothesized that during naturalistic encoding, at critical visibility ranges, higher contrast images will lead to higher signal-to-noise ratio and better signal quality flowing downstream and will thus be better remembered. Indeed, we found that during naturalistic encoding higher contrast images were remembered better than lower contrast ones (~ 15% higher accuracy, ~ 1.58 times better) for images at 7.5–60 RMS contrast range. Although image contrast and size modulate early visual processing very differently, our results further substantiate that at poor visibility ranges, during naturalistic non-instructed visual behavior, physical image dimensions (contributing to image visibility) impact image memory.

Improved Identification of Forest Types in the Loess Plateau Using Multi-Source Remote Sensing Data, Transfer Learning, and Neural Residual Networks

This study aims to establish a deep learning-based classification framework to efficiently and rapidly distinguish between coniferous and broadleaf forests across the Loess Plateau. By integrating the deep residual neural network (ResNet) architecture with transfer learning techniques and multispectral data from unmanned aerial vehicles (UAVs) and Landsat remote sensing data, the effectiveness of the framework was validated through well-designed experiments. The study began by selecting optimal spectral band combinations, using the random forest algorithm. Pre-trained models were then constructed, and model performance was optimized with different training strategies, considering factors such as image size, sample quantity, and model depth. The results indicated substantial improvements in the model’s classification accuracy and efficiency for reasonable image dimensions and sample sizes, especially for an image size of 3 × 3 pixels and 2000 samples. In addition, the application of transfer learning and model fine-tuning strategies greatly enhanced the adaptability and universality of the model in different classification scenarios. The fine-tuned model achieved remarkable performance improvements in forest-type classification tasks, increasing classification accuracy from 85% to 93% in Zhengning, from 89% to 96% in Yongshou, and from 86% to 94% in Baishui, as well as exceeding 90% in all counties. These results not only confirm the effectiveness of the proposed framework, but also emphasize the roles of image size, sample quantity, and model depth in improving the generalization ability and classification accuracy of the model. In conclusion, this research has developed a technological framework for effective forest landscape recognition, using a combination of multispectral data from UAVs and Landsat satellites. This combination proved to be more effective in identifying forest types than was using Landsat data alone, demonstrating the enhanced capability and accuracy gained by integrating UAV technology. This research provides valuable scientific guidance and tools for policymakers and practitioners in forest management and sustainable development.

Multi-detector fusion and Bayesian smoothing for tracking viral and chromatin structures

Automatic tracking of viral and intracellular structures displayed as spots with varying sizes in fluorescence microscopy images is an important task to quantify cellular processes. We propose a novel probabilistic tracking approach for multiple particle tracking based on multi-detector and multi-scale data fusion as well as Bayesian smoothing. The approach integrates results from multiple detectors using a novel intensity-based covariance intersection method which takes into account information about the image intensities, positions, and uncertainties. The method ensures a consistent estimate of multiple fused particle detections and does not require an optimization step. Our probabilistic tracking approach performs data fusion of detections from classical and deep learning methods as well as exploits single-scale and multi-scale detections. In addition, we use Bayesian smoothing to fuse information of predictions from both past and future time points. We evaluated our approach using image data of the Particle Tracking Challenge and achieved state-of-the-art results or outperformed previous methods. Our method was also assessed on challenging live cell fluorescence microscopy image data of viral and cellular proteins expressed in hepatitis C virus-infected cells and chromatin structures in non-infected cells, acquired at different spatial–temporal resolutions. We found that the proposed approach outperforms existing methods.

MultiTrans: Multi-branch transformer network for medical image segmentation

Background and Objective:Transformer, which is notable for its ability of global context modeling, has been used to remedy the shortcomings of Convolutional neural networks (CNN) and break its dominance in medical image segmentation. However, the self-attention module is both memory and computational inefficient, so many methods have to build their Transformer branch upon largely downsampled feature maps or adopt the tokenized image patches to fit their model into accessible GPUs. This patch-wise operation restricts the network in extracting pixel-level intrinsic structural or dependencies inside each patch, hurting the performance of pixel-level classification tasks. Methods:To tackle these issues, we propose a memory- and computation-efficient self-attention module to enable reasoning on relatively high-resolution features, promoting the efficiency of learning global information while effective grasping fine spatial details. Furthermore, we design a novel Multi-Branch Transformer (MultiTrans) architecture to provide hierarchical features for handling objects with variable shapes and sizes in medical images. By building four parallel Transformer branches on different levels of CNN, our hybrid network aggregates both multi-scale global contexts and multi-scale local features. Results:MultiTrans achieves the highest segmentation accuracy on three medical image datasets with different modalities: Synapse, ACDC and M&Ms. Compared to the Standard Self-Attention (SSA), the proposed Efficient Self-Attention (ESA) can largely reduce the training memory and computational complexity while even slightly improve the accuracy. Specifically, the training memory cost, FLOPs and Params of our ESA are 18.77%, 20.68% and 74.07% of the SSA. Conclusions:Experiments on three medical image datasets demonstrate the generality and robustness of the designed network. The ablation study shows the efficiency and effectiveness of our proposed ESA. Code is available at: https://github.com/Yanhua-Zhang/MultiTrans-extension.

Investigation of Appropriate Scaling of Networks and Images for Convolutional Neural Network-Based Nerve Detection in Ultrasound-Guided Nerve Blocks.

Ultrasound imaging is an essential tool in anesthesiology, particularly for ultrasound-guided peripheral nerve blocks (US-PNBs). However, challenges such as speckle noise, acoustic shadows, and variability in nerve appearance complicate the accurate localization of nerve tissues. To address this issue, this study introduces a deep convolutional neural network (DCNN), specifically Scaled-YOLOv4, and investigates an appropriate network model and input image scaling for nerve detection on ultrasound images. Utilizing two datasets, a public dataset and an original dataset, we evaluated the effects of model scale and input image size on detection performance. Our findings reveal that smaller input images and larger model scales significantly improve detection accuracy. The optimal configuration of model size and input image size not only achieved high detection accuracy but also demonstrated real-time processing capabilities.

Improved dual-threshold quantum image segmentation algorithm and simulation* *This work was supported by National Natural Science Foundation of China (No.61772295), Key Projects of Chongqing Natural Science Foundation Innovation Development Joint Fund (CSTB2023NSCQ-LZX0139), Open Fund of Advanced Cryptography and System Security Key Laboratory of Sichuan Province (Grant No. SKLACSS-202208) and Technology Research Program of Chongqing Municipal Education Commission (Grant no. KJZD-M202000501).

Quantum image segmentation algorithm is crucial for quantum image processing. In this paper, a dual-threshold quantum image segmentation algorithm is designed and simulated in IBM Quantum Experience (IBM Q) platform, which can segment a complex image into three parts using fewer quantum bits. In our algorithm, given a high threshold and a low threshold, grayscale values larger than the high threshold are set to the high threshold and grayscale values smaller than the low threshold are set to the low threshold, with no change for the part between the two thresholds. Then we use a low-cost quantum comparator and design a complete and scalable quantum image segmentation circuit. Analysis of the circuit cost shows that the quantum gates required for the circuit are only related to the grayscale range q and are independent of the image size. The feasibility of the algorithm and the correctness of the quantum circuit are verified by simulation in IBM Q platform, and finally the MSE, PSNR AND SSIM value of the image is analyzed to prove the effectiveness of the segmentation algorithm.

Discrete Curvelet Transform Feature Extraction for Mangosteen Fruit Surface Damage Detection

Mangosteen (Garcinia mangostana L) is one of the commodities of Indonesian fruit and is used as an export primadona that became the basis of Indonesia to increase the currency of the country. The quality of the fruit can be seen from the surface, whether there is damage or not. The sorting that the farmers have been doing all this time is still using the conventional way, that is, with the sense of sight. This conventional method seems to be less effective because it takes a lot of energy, takes a long time, and there are different perceptions between farmers. To solve this problem, a method of surface quality extraction of mango fruit will be developed based on image processing. The initial stage of image processing is with the image size equation then the image is converted to grayscale mode, then a discrete curvelet transformation is performed. The next stage is the extraction of mean, energy, entropy, standard deviation, variance, sum, correlation, contrast, and homogeneity. The result of the subsequent feature extraction is used to enter a value at the classification stage. From some of these extractions it will be known which extraction has the highest accuracy value. The method of classification used is Linear Discriminant Analysis (LDA) with the method of K-Fold Cross Validation which in this study is divided into 4-fold cross validation. After testing on 120 images, the highest value of accuracy is with extraction of standard characteristics deviation of 91.7% and variance of 88.4%.