Low-frequency Image Research Articles

Retrieval of Interior Structure of Asteroids with the Low-Frequency Telescope DART

Understanding the internal structure of asteroids is crucial for deciphering their formation and establishing defenses against potential hazards. The Daocheng Radio Telescope (DART), a recently constructed interferometric array designed for low-frequency Sun imaging, presents a promising tool for probing asteroid interiors. With a substantial 1-km array aperture and an equivalent receiving area of approximately 8,850 m 2 , DART plays a vital role in diagnosing asteroid internal structures. This study introduces an electromagnetic wave scattering model tailored to asteroids within DART’s operational frequency range (150 to 450 MHz). Ground-based radar detection can unveil multiple facets of these celestial bodies by leveraging low-frequency waves’ penetrating capabilities and capitalizing on asteroids’ rotational dynamics. Through simulations capturing the characteristics of low-frequency waves traversing a layered model and interacting with internal structures, we propose an electromagnetic scattering model of asteroids. Our results underscore DART’s potential as a crucial instrument for discerning the internal structure of near-Earth objects. We first formulate an asteroid model through celestial impact models, dimensional analysis, and data fitting to achieve this. Subsequently, we derive an electromagnetic scattering model using geometric optics and a propagation model for lossy mediums. Simulations demonstrate that morphology and internal structure dictate the distribution of scattered waves, with forward and backscattered waves providing comprehensive internal structure information over a rotation cycle. Furthermore, we observe that alterations in electromagnetic wave frequency induce changes in the scattering characteristics, prompting the convenience of employing multiple frequencies for retrieving detailed information about an asteroid’s internal medium and structure. This multidimensional approach positions DART as a promising asset in advancing our understanding of asteroid interiors, offering valuable insights for scientific inquiry and hazard mitigation strategies.

Foliage-Concealed Target Change Detection Scheme Based on Convolutional Neural Network in Low-Frequency Ultrawideband SAR Images

Foliage-Concealed Target Change Detection Scheme Based on Convolutional Neural Network in Low-Frequency Ultrawideband SAR Images

Attenuation Tomography Using Low-frequency Ultrasound with Variational Autoencoder for Thorax Imaging: Experimental Study.

In this paper, we introduce a novel inversion methodology employing the variational autoencoder (VAE) for human thorax attenuation tomography using low-frequency ultrasound. The VAE is trained to assimilate the structural priors of the human thorax, utilizing training samples generated from computed tomography (CT) scans. This approach enables the compression of high-dimensional attenuation distributions into a lowerdimensional latent space. During the inversion process, the latent code is optimized, and then the reconstructed model is generated by the decoder of the VAE. This process can effectively integrate prior information of the domain of interest (DOI) into the inversion through coding and decoding, which would mitigate the ill-posedness of the inverse problem and facilitate better outcomes. Our method demonstrates robust generalization capabilities and noise resilience in numerical simulations, outperforming the conventional pixel-based Gauss-Newton method. Human subject experiment further corroborates the effectiveness of our approach. This is also the first experimental validation of the feasibility of low-frequency ultrasound functional imaging of the human thorax. Although the current study presents certain limitations, it underscores the potential of low-frequency ultrasound in the continuous monitoring of the human respiratory system.

Wavelet Pyramid Recurrent Structure-Preserving Attention Network for Single Image Super-Resolution.

Many single image super-resolution (SISR) methods that use convolutional neural networks (CNNs) learn the relationship between low-and high-resolution images directly, without considering the context structure and detail fidelity. This can limit the potential of CNNs and result in unrealistic, distorted edges and textures in the reconstructed images. A more effective approach is to incorporate prior knowledge about the image into the model to aid in image reconstruction. In this study, we propose a novel recurrent structure-preserving mechanism that innovatively uses the multiscale wavelet transform (WT) as an image prior, namely, wavelet pyramid recurrent structure-preserving attention network (WRSANet), to process both low-and high-frequency subnetworks at each level separately and recursively. We propose a novel structure scale preservation (SSP) architecture that differs from traditional WTs. This architecture allows us to incorporate and learn structure preservation subnetworks at each level. By using our proposed structure scale fusion (SSF) combined with inverse WT, we can recursively restore and preserve rich low-frequency image structure through the combination of SSP at various levels. Furthermore, we also propose novel low-to-high-frequency information transmission (L2HIT) and detail enhancement (DE) mechanisms to address the issue of detail distortion in high-frequency images by transferring information from structure preservation subnetworks. This allows us to preserve the low-frequency structure while reconstructing high-frequency details, improving detail fidelity and avoiding structural distortion. Finally, a joint loss function is also used to balance the fusion of low-and high-frequency information at different degrees, with hyperparameters being adjusted during training. The experimental results demonstrate that the proposed WRSANet achieves better performance and visual presentation than the state-of-the-art (SOTA) on synthetic and real datasets, especially in terms of context structure and texture details.

Ultrasound imaging findings in primary biliary cholangitis

BackgroundOur study aimed to analyze the characteristics of ultrasound images corresponding to each histological stage of primary biliary cholangitis (PBC).MethodsWe prospectively analyzed 75 confirmed cases of PBC and used liver biopsy as the gold standard to determine the disease stage.ResultsThe typical ultrasound images of patients with PBC were characterized by a thickening of the portal vein wall (PVW) and periportal hypoechoic band (PHB) width with increasing histological stages, and significant increases in the left hepatic lobe diameter (LHLD) in stage II (by 64.0%) and stage III (by 69.2%). PHB width (r = 0.857, p < 0.001), PVW thickness (r = 0.488, p < 0.001), and spleen area (r = 0.8774, p < 0.001) were positively correlated with the histological stage. Significant changes were noted in the liver surface, echo texture, and edge between different stages. The areas under the receiver operating characteristic curve of composite indicators were 0.965 for predicting progressive PBC(≥ stage 2), and 0.926 for predicting advanced PBC(≥ stage 3).ConclusionsThe ultrasound imaging characteristics of patients with PBC varied according to the histological staging. LHLD, PVW thickness, and PHB width were significantly correlated with the histological stage. A combination of high- and low-frequency ultrasound imaging can provide relevant cues regarding the degree of PBC progression and important clinical reference values. The application of all the ultrasound image findings as the composite indicators can better predict progressive and advanced PBC, providing important clinical reference values.

Investigating the spontaneous brain activities of patients with subjective cognitive decline and mild cognitive impairment: an amplitude of low-frequency fluctuation functional magnetic resonance imaging study.

Subjective cognitive decline (SCD) and mild cognitive impairment (MCI) are neurodegenerative processing stages of Alzheimer's disease (AD). Cognitive decline is thought to manifest in intrinsic brain activity changes, but research results yielded conflicting and few studies have explored the roles of brain regions in cognitive decline, and sensitivity of the cognitive field to changes in the altered intrinsic brain activity. In this cross-sectional study, 158 elderly participants were recruited from the memory clinic of the First Affiliated Hospital of Nanjing Medical University from July 2019 to May 2021, and grouped into SCD (n=73), MCI (n=46), and normal controls (NC) (n=39). The amplitude of low-frequency fluctuation (ALFF) was calculated and evaluated among the groups. Then canonical correlation analysis (CCA) was conducted to investigate the associations between imaging outcomes and cognitive behaviors. Neuropsychological tests in different cognitive dimensions and ALFF values of the prefrontal, parietal, and temporal gyrus, were significantly different (P<0.05) among the three groups, with no appreciable decline in daily activity. The changes in intrinsic activities were closely related to the decline in cognitive function (R=0.73, P=0.002). ALFF values in the left middle occipital gyrus, right middle frontal gyrus, left superior frontal gyrus, left angular gyrus, and superior temporal gyrus played significant roles in the analysis, while the Montreal Cognitive Assessment (MoCA) and Auditory-Verbal Learning Test scores were found to be more sensitive to changes in ALFF values. Spontaneous brain activity is a stable imaging biomarker of cognitive impairment. ALFF changes of the prefrontal, occipital, left angular, and temporal gyrus were sensitive to identifying cognitive decline, and the scores of the Auditory-Verbal Learning Test and MoCA could predict the abnormal intrinsic activities.

Improvement of low-frequency ultrasonic image quality using a enhanced convolutional neural network

Ultrasound has become an indispensable clinical scanning and diagnostic tool because of its convenience, rapidity, and radiation-free properties. In ultrasound imaging, the resolution and detection depth of ultrasound images are closely related to the frequency of the transmitted ultrasound. Ultrasound with low transmission frequency has a wide detection range, but its images have problems such as low resolution, blurred edge contours and obvious background noise. In contrast, the higher transmission frequency, the smaller the detection range, but the resolution has improved. The rapid development of deep learning provides new approaches and enormous potential for improving resolution of ultrasound images and acquiring high-resolution images with an extensive detection range. In this paper, the attention mechanism is combined into a convolutional neural network based on the encoder-decoder, and the feature maps extracted at different scales are merged by skip connections to preserve more structure and contrast details, thus realizing the improvement of lower-frequency ultrasound image resolution. This work utilizes the capacitive micromechanical ultrasound transducer (CMUT) probe for dataset acquisition. The network's performance is validated by simulated datasets for point targets and experimental datasets for breast body models. The experimental results show that for ultrasound images, the network output map can improve by 0.2 in structural similarity and at least 11 dB in peak signal-to-noise ratio. The proposed network can effectively preserve ultrasound images' structural and textural information, improve edge contour definition, and significantly suppress noise and imaging artifacts, which has significant development potential in ultrasound imaging.

Two-branch crisscross network for realistic and accurate image super-resolution

Convolutional Neural Networks have made remarkable progress in single-image super-resolution. However, existing methods struggle to balance reconstruction accuracy and perceptual quality, resulting in unsatisfactory outcomes. To address this challenge, we propose the Two-Branch Crisscross Generative Adversarial Network (TBCGAN) for achieving accurate and realistic super-resolution results. TBCGAN employs two asymmetric branches that separately reconstruct high-frequency (HF) and low-frequency (LF) images, leveraging their distinct information and reconstruction requirements. To ensure coherent results, we apply different supervision to the reconstructed HF, LF, and super-resolution (SR) images while facilitating information interaction through the interleaving and fusion of HF and LF features. Extensive experimental evaluations demonstrate that TBCGAN achieves an excellent balance between reconstruction accuracy and perceptual quality, outperforming GAN-based methods in reconstruction accuracy and MSE-based methods in perceptual quality.

Remote Sensing Image Super-Resolution Using Texture Enhancing Generative Adversarial Network

&lt;p&gt;Single image super-resolution (SISR) brings excellent improvement in remote sensing applications, which has been widely studied in recent years. A method named TFSRGAN of remote sensing single image super-resolution based on generative adversarial network is proposed in this paper to address the problems of poor reconstruction visual quality and smooth details in traditional algorithms. In the proposed framework, s dense residual connection method is proposed to fuse the deep features from each residual block based on the SRGAN network, and the channel attention mechanism is added into the residual block to combinate the channel information. In addition, the network employs an edge extractor to divide the low-resolution image into low-frequency image and high-frequency image as the input of generator to improve the effect of texture reconstruction. Extensive comparison experiments were performed using AID, UCAS_AOD and China Gaofen-1 datasets, the SR results demonstrate that the proposed TFSRGAN framework outperforms the state-of-the-art algorithms including VDSR, SRGAN and ESRGAN in terms of objective evaluation metrics and subjective visual perception. The ground targets detection experiments represent that the proposed TFSRGAN can significantly improve the effect in remote sensing super-resolution application.&lt;/p&gt; &lt;p&gt;&amp;nbsp;&lt;/p&gt;

Illusory optical defocus generated by shaded surface texture

The human visual system is tasked with the problem of extracting information about the world from images that contain a conflated mixture of environmental sources and optical artifacts generated by the focal properties of our eyes. In most contexts, our brains manage to distinguish these sources, but this is not always the case. Recent work showed that shading gradients generated by smooth three-dimensional (3D) surfaces can elicit strong illusory percepts of optical defocus1,2 - the perception of illusory blur is only eliminated when the surface appears attached to self-occluding contours3, surface discontinuities1, or sharp specular reflections1,2, which all generate sharp ('high spatial frequency') image structure. This suggests that it should also be possible to eliminate the illusory blur elicited by shaded surfaces by altering the surface geometry to include small-scale surface relief, which would also generate high-frequency image structure. We report the surprising result here that this manipulation fails to eliminate the perception of blur; the fine texture fails to perceptually 'bind' to the low-frequency image structure when there is a sufficient gap between the spatial scales of the fine and coarse surface structure. These findings suggest that discontinuous 'gaps' in the spatial scale of textures are a segmentation cue the visual system uses to extract multiple causes of image structure.

Laplacian Pyramid Fusion Network With Hierarchical Guidance for Infrared and Visible Image Fusion

The fusion of infrared and visible images combines the information from two complementary imaging modalities for various computer vision tasks. Many existing techniques, however, fail to maintain a uniform overall style and keep salient details of individual modalities simultaneously. This paper presents an end-to-end Laplacian Pyramid Fusion Network with hierarchical guidance (HG-LPFN) that takes advantage of pixel-level saliency reservation of Laplacian Pyramid and global optimization capability of deep learning. The proposed scheme generates hierarchical saliency maps through Laplacian Pyramid decomposition and modal difference calculation. In the pyramid fusion mode, all sub-networks are connected in a bottom-up manner. The sub-network for low-frequency fusion focuses on extracting universal features to produce an opposite style while sub-networks for high-frequency fusion determine how much the details of each modality will be retained. Taking the style, details, and background into consideration, we design a set of novel loss functions to supervise both low-frequency images and full-resolution images under the guidance of saliency maps. Experimental results on public datasets demonstrate that the proposed HG-LPFN outperforms the state-of-the-art image fusion techniques.

An efficient image defogging algorithm based on dragonfly optimized gamma correction and stationary wavelet decomposition

Image dehazing is a revolutionary technique for restoring images with hazy or foggy landscapes, that has gotten a lot of focus in recent years since it gained importance in a surveillance system. However, the image processing by the traditional defogging algorithm has difficulties in integrating the depth of image detail and the color of the image. Therefore, in this paper, a novel framework based on wavelet decomposition and optimized gamma correction is proposed for efficaciously retrieving the fog-free image. The foggy image is first divided into low and high frequency sub-images using SWT (Stationary Wavelet Transform), which has the advantages of preserving temporal features so that information loss can be stopped. Then the low frequency and high frequency images are processed with defogging and denoising modules to remove fog and noise respectively. The DOGC (Dragonfly optimal Gamma Correction) algorithm in dehazing module dynamically enhanced the color detail information without human intervention so that observed scene contrast and visibility are well preserved. Lastly, fog-free image is reconstructed from sub-enhanced images. The experimental findings show that the proposed framework outperforms state-of-the-art methods in terms of both quantitative and qualitative assessment criteria using the established dataset. Furthermore, the proposed method efficiently removes fog while preserving the naturalness of fog images.

An Unsupervised Machine Learning-based Algorithm for Detecting Weak Impulsive Narrowband Quiet Sun Emissions and Characterizing Their Morphology

The solar corona is extremely dynamic. Every leap in observational capabilities has been accompanied by unexpected revelations of complex dynamic processes. The ever more sensitive instruments now allow us to probe events with increasingly weaker energetics. A recent leap in the low-frequency radio solar imaging ability has led to the discovery of a new class of emissions, namely weak impulsive narrowband quiet Sun emissions (WINQSEs). They are hypothesized to be the radio signatures of coronal nanoflares and could potentially have a bearing on the long standing coronal heating problem. In view of the significance of this discovery, this work has been followed up by multiple independent studies. These include detecting WINQSEs in multiple data sets, using independent detection techniques and software pipelines, and looking for their counterparts at other wavelengths. This work focuses on investigating morphological properties of WINQSEs and also improves upon the methodology used for detecting WINQSEs in earlier works. We present a machine learning-based algorithm to detect WINQSEs, classify them based on their morphology, and model the isolated ones using 2D Gaussians. We subject multiple data sets to this algorithm to test its veracity. Interestingly, despite the expectations of their arising from intrinsically compact sources, WINQSEs tend to be resolved in our observations. We propose that this angular broadening arises due to coronal scattering. Hence, WINQSEs can provide ubiquitous and ever-present diagnostic of coronal scattering (and, in turn, coronal turbulence) in the quiet Sun regions, which has not been possible until date.

Deep Learning-Based Kernel Adaptation Enhances Quantification of Emphysema on Low-Dose Chest CT for Predicting Long-Term Mortality.

The aim of this study was to ascertain the predictive value of quantifying emphysema using low-dose computed tomography (LDCT) post deep learning-based kernel adaptation on long-term mortality. This retrospective study investigated LDCTs obtained from asymptomatic individuals aged 60 years or older during health checkups between February 2009 and December 2016. These LDCTs were reconstructed using a 1- or 1.25-mm slice thickness alongside high-frequency kernels. A deep learning algorithm, capable of generating CT images that resemble standard-dose and low-frequency kernel images, was applied to these LDCTs. To quantify emphysema, the lung volume percentage with an attenuation value less than or equal to -950 Hounsfield units (LAA-950) was gauged before and after kernel adaptation. Low-dose chest CTs with LAA-950 exceeding 6% were deemed emphysema-positive according to the Fleischner Society statement. Survival data were sourced from the National Registry Database at the close of 2021. The risk of nonaccidental death, excluding causes such as injury or poisoning, was explored according to the emphysema quantification results using multivariate Cox proportional hazards models. The study comprised 5178 participants (mean age ± SD, 66 ± 3 years; 3110 males). The median LAA-950 (18.2% vs 2.6%) and the proportion of LDCTs with LAA-950 exceeding 6% (96.3% vs 39.3%) saw a significant decline after kernel adaptation. There was no association between emphysema quantification before kernel adaptation and the risk of nonaccidental death. Nevertheless, after kernel adaptation, higher LAA-950 (hazards ratio for 1% increase, 1.01; P = 0.045) and LAA-950 exceeding 6% (hazards ratio, 1.36; P = 0.008) emerged as independent predictors of nonaccidental death, upon adjusting for age, sex, and smoking status. The application of deep learning for kernel adaptation proves instrumental in quantifying pulmonary emphysema on LDCTs, establishing itself as a potential predictive tool for long-term nonaccidental mortality in asymptomatic individuals.

Speckle Noise Reduction in Ultrasound Images

Abstract: Ultrasound Imaging is used to examine various organs in the Human Body. However, in the process of obtaining a Ultrasound Image, Speckle noise is generated due to backscattered echo signals. So, a noise reduction method is required. To improve the speckle noise reduction, we propose a novel algorithm using characteristics of speckle noise and filtering methods based on Speckle Reducing Anisotropic Diffusion (SRAD) filtering, Discrete Wavelet Transform (DWT), Weighted Guided Image Filtering (WGIF) and Gradient Domain Guided Image Filtering (GDGIF). The SRAD filter is exploited as a preprocessing filter because it can be directly applied to a medical US image containing speckle noise without a log-compression. The wavelet domain has the advantage of suppressing the additive noise. Therefore, a homomorphic transformation is utilized to convert the multiplicative noise into additive noise. After two-level DWT decomposition is applied, to suppress the residual noise of an SRAD filtered image, GDGIF and WGIF are exploited to reduce noise from seven high-frequency subband images and one low-frequency sub-band image, respectively. Finally, a noise-free image is attained through inverse DWT and an exponential transform

Low-Frequency ultrasonic tomography of Corrosion-induced damage patterns on naturally corroded solid reinforcing bar rock bolts

A combined non-destructive guided ultrasonic wave measuring approach and low-frequency ultrasonic tomography technique on cylindrical surfaces is developed to investigate corrosion-induced damage patterns in naturally corroded 60-year-old solid rebars. The main goal of low-frequency tomographic imaging simulations is for differential detection of general and localized pitting corrosions on cylindrical surfaces of solid steel bars. The modified damage probabilistic reconstruction algorithm for cylindrical surfaces is also adopted to simulate naturally general and localized pitting corrosion patterns. To increase the ray density or resolution of images, a regular array of accelerometer sensors with high-density spatial distributions around the cylindrical surface of steel bars is installed for a fixed piezocomposite actuator. As such, an efficient sensor deployment method can be applied with the trilateration technique to achieve effective coverage and connectivity properties between accelerometer sensors. Stochastic signal processing which is elicited from the Hilbert-Huang Transform (HHT) signal method is also employed to analysis low-frequency tomography time domain algorithms. Finally, optical surface profilometry approaches, and static state tensile testings, together with scanning electron microscopy (SEM) are adopted to confirm the effectiveness of non-destructive and tomography imaging techniques. The results show that the guided wave signals with excitation frequency of less than 10 kHz can be precisely employed to identify the general and localized pitting corrosion distributions on the cylindrical surfaces of steel bars and qualitatively reconstruct them using low-frequency ultrasonic tomography simulations. This research study also demonstrates that the low-frequency ultrasonic tomography simulations can be effectively applied in corrosion pattern uncertainty modelling.

Estimation of brain tissue response by electrical stimulation in a subject-specific model implemented by conductivity tensor imaging.

Electrical stimulation such as transcranial direct current stimulation (tDCS) is widely used to treat neuropsychiatric diseases and neurological disorders. Computational modeling is an important approach to understand the mechanisms underlying tDCS and optimize treatment planning. When applying computational modeling to treatment planning, uncertainties exist due to insufficient conductivity information inside the brain. In this feasibility study, we performed in vivo MR-based conductivity tensor imaging (CTI) experiments on the entire brain to precisely estimate the tissue response to the electrical stimulation. A recent CTI method was applied to obtain low-frequency conductivity tensor images. Subject-specific three-dimensional finite element models (FEMs) of the head were implemented by segmenting anatomical MR images and integrating a conductivity tensor distribution. The electric field and current density of brain tissues following electrical stimulation were calculated using a conductivity tensor-based model and compared to results using an isotropic conductivity model from literature values. The current density by the conductivity tensor was different from the isotropic conductivity model, with an average relative difference |rD| of 52 to 73%, respectively, across two normal volunteers. When applied to two tDCS electrode montages of C3-FP2 and F4-F3, the current density showed a focused distribution with high signal intensity which is consistent with the current flowing from the anode to the cathode electrodes through the white matter. The gray matter tended to carry larger amounts of current densities regardless of directional information. We suggest this CTI-based subject-specific model can provide detailed information on tissue responses for personalized tDCS treatment planning.

CT and MRI fusion based on generative adversarial network and convolutional neural networks under image enhancement

Aiming at the problems of missing important features, inconspicuous details and unclear textures in the fusion of multimodal medical images, this paper proposes a method of computed tomography (CT) image and magnetic resonance imaging (MRI) image fusion using generative adversarial network (GAN) and convolutional neural network (CNN) under image enhancement. The generator aimed at high-frequency feature images and used double discriminators to target the fusion images after inverse transform; Then high-frequency feature images were fused by trained GAN model, and low-frequency feature images were fused by CNN pre-training model based on transfer learning. Experimental results showed that, compared with the current advanced fusion algorithm, the proposed method had more abundant texture details and clearer contour edge information in subjective representation. In the evaluation of objective indicators, Q AB/F, information entropy (IE), spatial frequency (SF), structural similarity (SSIM), mutual information (MI) and visual information fidelity for fusion (VIFF) were 2.0%, 6.3%, 7.0%, 5.5%, 9.0% and 3.3% higher than the best test results, respectively. The fused image can be effectively applied to medical diagnosis to further improve the diagnostic efficiency.

Denoising Diffusion Probabilistic Feature-Based Network for Cloud Removal in Sentinel-2 Imagery

Cloud contamination is a common issue that severely reduces the quality of optical satellite images in remote sensing fields. With the rapid development of deep learning technology, cloud contamination is expected to be addressed. In this paper, we propose Denoising Diffusion Probabilistic Model-Cloud Removal (DDPM-CR), a novel cloud removal network that can effectively remove both thin and thick clouds in optical image scenes. Our network leverages the denoising diffusion probabilistic model (DDPM) architecture to integrate both clouded optical and auxiliary SAR images as input to extract DDPM features, providing significant information for missing information retrieval. Additionally, we propose a cloud removal head adopting the DDPM features with an attention mechanism at multiple scales to remove clouds. To achieve better network performance, we propose a cloud-oriented loss that considers both high- and low-frequency image information as well as cloud regions in the training procedure. Our ablation and comparative experiments demonstrate that the DDPM-CR network outperforms other methods under various cloud conditions, achieving better visual effects and accuracy metrics (MAE = 0.0229, RMSE = 0.0268, PSNR = 31.7712, and SSIM = 0.9033). These results suggest that the DDPM-CR network is a promising solution for retrieving missing information in either thin or thick cloud-covered regions, especially when using auxiliary information such as SAR data.

Facilitating deep learning through preprocessing of optical coherence tomography images

BackgroundWhile deep learning has delivered promising results in the field of ophthalmology, the hurdle to complete a deep learning study is high. In this study, we aim to facilitate small scale model trainings by exploring the role of preprocessing to reduce computational burden and accelerate learning.MethodsA small subset of a previously published dataset containing optical coherence tomography images of choroidal neovascularization, drusen, diabetic macula edema, and normal macula was modified using Fourier transformation and bandpass filter, producing high frequency images, original images, and low frequency images. Each set of images was trained with the same model, and their performances were compared.ResultsCompared to that with the original image dataset, the model trained with the high frequency image dataset achieved an improved final performance and reached maximum performance much earlier (in fewer epochs). The model trained with low frequency images did not achieve a meaningful performance.ConclusionAppropriate preprocessing of training images can accelerate the training process and can potentially facilitate modeling using artificial intelligence when limited by sample size or computational power.