A Perceptual Encryption-Based Image Communication System for Deep Learning-Based Tuberculosis Diagnosis Using Healthcare Cloud Services

Segmentation of color images is a tricky task. Regaining the segments in an image using the content image is a difficult and significant problem. Medical imaging is the most alive research topic from the past two decades. From the medical diagnosis of patients suffering from various diseases, abnormal regions in the organs can be easily identified, which is a greatest achievement. It is experimentally proved that graph based segmentation methods are better than the other segmentation techniques, especially when combined with statistical methods. We have proposed color image segmentation by adaptive graph cut method in this paper. It consists of two important stages. During the first phase we enhance the input color image using transformation technique as the image may contain noise, may be of low contrast and missing some color statistics. Then this enhanced color image is processed under graph cut technique to get better results, especially for the analysis of medical and general images. The proposed method contributes to medical imaging by means of image segmentation and also to other general image analysis. Our experimental results are found to very good in segmenting color images.

Radiologists' eye movements during medical image interpretation reflect their perceptual-cognitive processes of diagnostic decisions. The eye movement data can be modeled to represent clinically relevant regions in a medical image and potentially integrated into an artificial intelligence (AI) system for automatic diagnosis in medical imaging. In this article, we first conduct a large-scale eye-tracking study involving 13 radiologists interpreting 191 chest X-ray (CXR) images, establishing a best-of-its-kind CXR visual saliency benchmark. We then perform analysis to quantify the reliability and clinical relevance of saliency maps (SMs) generated for CXR images. We develop CXR image saliency prediction method (CXRSalNet), a novel saliency prediction model that leverages radiologists' gaze information to optimize the use of unlabeled CXR images, enhancing training and mitigating data scarcity. We also demonstrate the application of our CXR saliency model in enhancing the performance of AI-powered diagnostic imaging systems.

Most existing methods for edge detection rely on contrast dependent thresholds. We show that a local measurement defined by the ratio of the smallest to the largest eigenvalue of the second moment matrix of filter kernels, can be used to separate smooth, low curvature curves and straight lines from noise, independent of contrast, in both color and gray level images. This is done without applying a threshold to the gradient magnitude. The edge images are defined as zero crossings in the gradient direction. The covariance matrix can easily be computed for both gray level images and color images. Further we show the potentiality of such a measure by integrating it with the Hough transform to extract long straight lines in noisy color images. The method is shown to successfully extract consistent line features from color images of a scene, captured under drastically different lightening conditions.

Deep learning development nowadays has attracted a lot of attention because of its effectiveness and good performance. The performance of deep learning in medical images analysis already can compete with medical image experts. However, there are experts that still believe deep learning only efficient for the big datasets, because of deep learning performance in small datasets still not satisfying enough. In this study, it is aimed to build a deep learning model for image classification that can achieve high accuracy using chest X-ray images with a relatively small dataset. We classify chest X-ray into a binary classification which is a normal image and image with abnormalities. We built and experimented our model using the public dataset of Shenzen Hospital dataset. We also use a different type of input based on different images preprocessing so that the model can perform accurate classification. Based on the result, pre-trained CheXNet with a newly trained fully connected network on the cropped dataset can achieve the accuracy 0.8761, the sensitivity 0.8909, and the specificity 0.8621. The performance of the model also influenced by the certain region inside the images, such as other regions outside the lung region and black colored region outside the body region.

ATEDU-NET: An Attention-Embedded Deep Unet for multi-disease diagnosis in chest X-ray images, breast ultrasound, and retina fundus.

Steganography is the art of the hidden data in such a way that it detection of hidden knowledge prevents. As the necessity of security and privacy increases, the need of the hiding secret data is ongoing. In this paper proposed an enhanced detection of the 1-2-4 LSB steganography and RSA cryptography in Gray Scale and Color images. For color images, we apply 1-2-4 LSB on component of the RGB, then encrypt information applying RSA technique. For Gray Images, we use LSB to then encrypt information and also detect edges of gray image. In the experimental outcomes, calculate PSNR and MSE. We calculate peak signal noise ratio for quality and brightness. This method makes sure that the information has been encrypted before hiding it into an input image. If in any case the cipher text got revealed from the input image, the middle person other than receiver can't access the information as it is in encrypted form.

The widespread adoption of deep learning (DL) solutions in the healthcare organizations is obstructed by their compute intensive nature and dependability on massive datasets. In this regard, cloud-services such as cloud storage and computational resources are emerging as an effective solution. However, when the image data are outsourced to avail such services, there is a privacy concern that the data should be kept protected not only during transmission but during computations as well. To meet these requirements, this study proposed a privacy-preserving DL (PPDL) scheme that enable computations without the need of decryption. The encryption is based on perceptual encryption (PE) that only hides the perceivable information in an image while preserves other characteristics that are necessary for DL computations. Precisely, we have implemented a binary classifier based on EfficientNetV2 for the COVID-19 screening in the chest X-ray (CXR) images. For the PE algorithm, the suitability of two pixel-based and two block-based PE methods was analyzed. The analysis have shown that when global contents are left unmodified (pixel-based PE), then the DL-based model achieved the classification accuracy same as that of the plain images. On the other hand, for block-based PE algorithms, there is up to 3% drop in the model's accuracy and sensitivity scores.

Nowadays, the facilities that offers Internet in conjunction with the capabilities of the infrastructure of communication and information technologies, allowing that the multimedia data can be shared and transmitted in easy and efficient way. As consequence, the multimedia data can be copied or edited in an easy form, and this can be affecting its original content. In a context of digital images, this paper presents a tamper detection and localization algorithm in color images using secure block-based watermarking. This proposal is based on the frequency domain using the lifting wavelet transform, together with checksum criteria that make up the hash key codes, which they are hidden in the least significant bits of each low frequency component. The proposed method is evaluated in terms of imperceptibility and false positive, false negative and tamper detection rates, respectively. Watermark imperceptibility is measured using the peak signal to noise ratio and structural similarity index, respectively. The tamper detection ability has been tested on color and grayscale images using a wide variety of manipulations such as copy-paste, copymove, constant averaging, impulsive noise, and random manipulations. Comparative with the current state-of-the-art is provided.

Chest X-ray imaging is a crucial diagnostic tool in medical practice, used for assessing various respiratory, cardiovascular, and musculoskeletal conditions. Recent advances in machine learning and computer vision have led to the development of algorithms capable of analyzing chest X-ray images to identify signs of disease or abnormalities with high accuracy. This study focuses on the application of a customized WideResNet model, a deep learning architecture, to classify chest X-ray images quality and diagnose various diseases. Two datasets were used in this study: an in-distribution dataset containing 5000 chest X-ray images classified into four classes (Infiltration, Effusion, Atelectasis, and Mass) and an out-of-distribution dataset consisting of chest X-Rays images also classified into four classes (Cardiomegaly, Consolidation, Emphysema, No Findings). Both datasets were preprocessed using a Scharr filter to enhance image quality and reduce noise. The Scharr filter is known for its accuracy in detecting small and diagonal edges, making it a suitable choice for preprocessing chest X-ray images. The WideResNet model, an extension of the ResNet architecture, was employed for image classification. The model's architecture consists of multiple residual blocks stacked together, each containing convolutional layers, batch normalization layers, and ReLU activation functions. The skip connections within the residual blocks address the vanishing gradient problem, allowing the model to learn the identity function and preserve input information. Wide residual blocks, a key feature of the WideResNet model, were used to improve the network's capacity to learn complex features while reducing its depth. The customized WideResNet model was trained and tested on the in-distribution and the out-of-distribution dataset to evaluate its ability to generalize to other medical conditions beyond chest X-rays. The model's performance was assessed by fine-tuning hyperparameters such as network depth, residual block width, and the number of filters in each convolutional layer. This study demonstrates the potential of the WideResNet model, combined with effective preprocessing techniques, in improving the accuracy and speed of chest X-ray imaging for disease diagnosis. The application of deep learning models such as WideResNet in medical imaging can lead to better patient outcomes through faster and more accurate detection of diseases. Furthermore, the use of out-of-distribution datasets for testing showcases the model's robustness and potential applicability in various medical fields beyond chest X-ray imaging.

We propose a novel method to fuse true colour images with monochromatic non-visible range images that seeks to encode important structural information from monochromatic images efficiently but also preserve the natural appearance of the available true chromacity information. We utilise the β colour opponency channel of the lαβ colour as the domain to fuse information from the monochromatic input into the colour input by the way of robust grayscale fusion. This is followed by an effective gradient structure visualisation step that enhances the visibility of monochromatic information in the final colour fused image. Images fused using this method preserve their natural appearance and chromacity better than conventional methods while at the same time clearly encode structural information from the monochormatic input. This is demonstrated on a number of well-known true colour fusion examples and confirmed by the results of subjective trials on the data from several colour fusion scenarios.

Tone-mapping-operators (TMOs) provide a useful means for converting high dynamic range (HDR) images to low dynamic range (LDR) images so that they can be viewed on standard displays, but this may influence the visual quality of experience (QoE) of the end-user. There is a need to understand the impact of TMOs to inform the choice of TMO algorithms for different displays, especially for small-screen-devices (SSDs) such as those used in mobile phones. This is important, as mobile devices are becoming the primary means of consuming multimedia contents. However, few studies have been undertaken to assess the impact of TMOs and viewing devices (especially SSDs) on the visual QoE of the user when using. In this chapter, we evaluate subjectively and objectively, the commonly used TMOs in different displays and resolutions for colour and grey-scale HDR images. Our results show that viewing devices have an influence on the TMOs performance, suggesting the need for a careful choice of TMO to enhance the viewing-QoE of the end-user. As expected, the higher resolution, the better HDR-image quality. Surprisingly, there was no significant difference between the Mean of Opinion Score (MOS) scores for colour and grey-scale images in SSDs. The device and TMOs affect QoE for colour and grey HDR-image equally. We found Shannon entropy (SE) to be a good objective measure of quality for colour and grey HDR images, suggesting that entropy may find use in automated HDR quality control assessment schemes, while; HDR-VDP-2 is a good objective measure for colour HDR image only.

IEViT: An enhanced vision transformer architecture for chest X-ray image classification

Imperceptibility is one of the most important aspects of the steganography method. Specifically, for steganography in images, embedded messages must not be felt by the human visual system. Humans have a higher sensitivity in color images compared to grayscale images, therefore the method that is still being tested on grayscale images needs to be retested on color images. Previous research has shown that the bit flipping method can increase imperceptibility to around 9dB in grayscale images. In this research the bit flipping method is tested on color images in the RGB format, the message capacity embedded in the image is 1 bit per pixel. The results of the imperceptibility test turned out to result in a more varied increase due to the greater number of layers in the color image. However, the bit flipping method is also proven to work well on color images, and even a maximum PSNR increase of more than 13dB. At the extraction stage, the message image can also be extracted perfectly, namely with the value NC = 1.

Colour and greyscale (black and white) pictures look different to us, but it is not clear whether the difference in appearance is a consequence of the way our visual system uses colour signals or a by-product of our experience. In principle, colour images are qualitatively different from greyscale images because they make it possible to use different processing strategies. Colour signals provide important cues for segmenting the image into areas that represent different objects and for linking together areas that represent the same object. If this property of colour signals is exploited in visual processing we would expect colour stimuli to look different, as a class, from greyscale stimuli. We would also expect that adding colour signals to greyscale signals should change the way that those signals are processed. We have investigated these questions in behavioural and in physiological experiments. We find that male marmosets (all of which are dichromats) rapidly learn to distinguish between colour and greyscale copies of the same images. The discrimination transfers to new image pairs, to new colours and to image pairs in which the colour and greyscale images are spatially different. We find that, in a proportion of neurons recorded in the marmoset visual cortex, colour-shifts in opposite directions produce similar enhancements of the response to a luminance stimulus. We conclude that colour is, both behaviourally and physiologically, a distinctive property of images.

Biological diversity is decreasing at a rate of 100-1000 times pre-human rates [1] [2], and tropical rainforests are among the most vulnerable ecosystems. To avoid species extinction, we need to understand factors influencing the occurrence of species. Fast, reliable computer-assisted tools can help to describe the habitat and thus to understand species habitat associations. This understanding is of utmost importance for more targeted species conservation efforts. Due to logistical challenges and time-consuming manual processing of field data, months up to years are often needed to progress from data collection to data interpretation. Deep learning can be used to significantly shorten the time while keeping a similar level of accuracy. Here, we propose Habitat-Net: a novel Convolutional Neural Network (CNN) based method to segment habitat images of rainforests. Habitat-Net takes color images as input and after multiple layers of convolution and deconvolution produces a binary segmentation of an image. The primary contribution of Habitat-Net is the translation of medical imaging knowledge (inspired by U-Net [3]) to ecological problems. The entire Habitat-Net pipeline works automatically without any user interaction. Our only assumption is the availability of annotated images, from which Habitat-Net learns the most distinguishing features automatically. In our experiments, we use two habitat datasets: (1) canopy and (2) understory vegetation. We train the model with 800 canopy images and 700 understory images separately. Our testing dataset has 150 canopy and 170 understory images. We use the Dice coefficient and Jaccard Index to quantify the overlap between ground-truthed segmentation images and those obtained by Habitat-Net model. This results in a mean Dice Score (mean Jaccard Index) for the segmentation of canopy and understory images of 0.89 (0.81) and 0.79 (0.69), respectively. Compared to manual segmentation, Habitat-Net prediction is approximately 3K – 150K times faster. For a typical canopy dataset of 335 images, Habitat-Net reduces total processing time to 5 seconds (15 milliseconds/ image) from 4 hours (45 seconds/ image). In this study, we show that it is possible to speed up the data pipeline using deep learning in the ecological domain. In the future, we plan to create a freely available mobile app based on Habitat-Net technology to characterize the habitat directly and automated in the field. In combination with ecological models our tools will help to understand the ecology of some poorly known, but often highly threatened, species and thus contribute to more timely conservation interventions.
\nREFERENCES: 
\n1. Sachs et al. "Biodiversity conservation and the millennium development goals." Science 325.5947 (2009): 1502-1503.
\n2. Chapin Iii, F. Stuart, et al. "Consequences of changing biodiversity." Nature 405.6783 (2000): 234.
\n3. Ronneberger, Olaf, Philipp Fischer, and Thomas Brox. "U-net: Convolutional networks for biomedical image segmentation." International Conference on Medical image computing and computer-assisted intervention. Springer, Cham, 2015.