Diabetic retinopathy prediction based on vision transformer and modified capsule network.

Diabetic retinopathy is considered one of the most common diseases that can lead to blindness in the working age, and the chance of developing it increases as long as a person suffers from diabetes. Protecting the sight of the patient or decelerating the evolution of this disease depends on its early detection as well as identifying the exact levels of this pathology, which is done manually by ophthalmologists. This manual process is very consuming in terms of the time and experience of an expert ophthalmologist, which makes developing an automated method to aid in the diagnosis of diabetic retinopathy an essential and urgent need. In this paper, we aim to propose a new hybrid deep learning method based on a fine-tuning vision transformer and a modified capsule network for automatic diabetic retinopathy severity level prediction. The proposed approach consists of a new range of computer vision operations, including the power law transformation technique and the contrast-limiting adaptive histogram equalization technique in the preprocessing step. While the classification step builds up on a fine-tuning vision transformer, a modified capsule network, and a classification model combined with a classification model, The effectiveness of our approach was evaluated using four datasets, including the APTOS, Messidor-2, DDR, and EyePACS datasets, for the task of severity levels of diabetic retinopathy. We have attained excellent test accuracy scores on the four datasets, respectively: 88.18%, 87.78%, 80.36%, and 78.64%. Comparing our results with the state-of-the-art, we reached a better performance.

Multi-task approach based on combined CNN-transformer for efficient segmentation and classification of breast tumors in ultrasound images.

Accurate segmentation of breast ultrasound (BUS) images is crucial for early diagnosis and treatment of breast cancer. Further, the task of segmenting lesions in BUS images continues to pose significant challenges due to the limitations of convolutional neural networks (CNNs) in capturing long-range dependencies and obtaining global context information. Existing methods relying solely on CNNs have struggled to address these issues. Recently, ConvNeXts have emerged as a promising architecture for CNNs, while transformers have demonstrated outstanding performance in diverse computer vision tasks, including the analysis of medical images. In this paper, we propose a novel breast lesion segmentation network CS-Net that combines the strengths of ConvNeXt and Swin Transformer models to enhance the performance of the U-Net architecture. Our network operates on BUS images and adopts an end-to-end approach to perform segmentation. To address the limitations of CNNs, we design a hybrid encoder that incorporates modified ConvNeXt convolutions and Swin Transformer. Furthermore, to enhance capturing the spatial and channel attention in feature maps we incorporate the Coordinate Attention Module. Second, we design an Encoder-Decoder Features Fusion Module that facilitates the fusion of low-level features from the encoder with high-level semantic features from the decoder during the image reconstruction. Experimental results demonstrate the superiority of our network over state-of-the-art image segmentation methods for BUS lesions segmentation.

Watermarking approach based on Hermite transform and a sliding window algorithm.

Nowadays, the distribution of huge amounts of medical images through open networks in telemedicine applications has become increasingly faster and easier. Therefore, a number of considerations are introduced related to the risks of the illegal use of these images, as total diagnosis depends on them. Indeed, the patient's data management, storage, and transmission require a technique for boosting security, integrity and privacy measures in telehealthcare services. In fact, in our previous works, we used polynomial decompositions such as Chebychev orthogonal polynomial transform in medical image watermarking. We then customise our tools for finding the best candidate area for embedding the watermark, always seeking to provide the best solution to this issue. In this research, a variation to medical image watermarking approach based on Hermite transform (HT) is suggested for reliable management of medical data. In this approach, the HT is employed as a preprocessing step so as to extract the texture component of the host medical image. Afterward, a sliding window technique is done to select the most suitable regions for watermark embedding. Lastly, the Arnold transform is utilized for encrypting the watermark to strengthen the security of our scheme. Experiments were conducted on various modalities of medical images. Results indicate that the proposed scheme is robust when subjected to various attacks while preserving a high level of security and invisibility factors. Also, our method preserves the quality of medical images with a good embedding capacity. The obtained results support the use of Hermite Polynomials for the implementation of watermarking in the medical imaging context.

Diabetic Retinopathy Prediction Based on Wavelet Decomposition and Modified Capsule Network.

Diabetic retinopathy (DR) is one of the most common consequences of diabetes. It affects the retina, causing blood vessel damage which can lead to loss of vision. Saving patients from losing their sight or at least slowing the progress of this disease depends mainly on the early detection of this pathology, on top of the detection of its specific stage. Furthermore, the early detection of diabetic retinopathy and the follow-up of the patient's condition remains an arduous task, whether for an experienced expert ophthalmologist or a computer-aided diagnosis technician. In this paper, we aim to propose a new automatic diabetic retinopathy severity level detection method. The proposed approach merges the pyramid hierarchy of the discrete wavelet transform of the retina fundus image with the modified capsule network and the modified inception block proposed, in addition to a new deep hybrid model that concatenates the inception block with capsule networks. The performance of our proposed approach has been validated on the APTOS dataset, as it achieved a high training accuracy of 97.71% and a high testing accuracy score of 86.54%, which is considered one of the best scores achieved in this field using the same dataset.

Efficient U-Net Architecture with Multiple Encoders and Attention Mechanism Decoders for Brain Tumor Segmentation.

The brain is the center of human control and communication. Hence, it is very important to protect it and provide ideal conditions for it to function. Brain cancer remains one of the leading causes of death in the world, and the detection of malignant brain tumors is a priority in medical image segmentation. The brain tumor segmentation task aims to identify the pixels that belong to the abnormal areas when compared to normal tissue. Deep learning has shown in recent years its power to solve this problem, especially the U-Net-like architectures. In this paper, we proposed an efficient U-Net architecture with three different encoders: VGG-19, ResNet50, and MobileNetV2. This is based on transfer learning followed by a bidirectional features pyramid network applied to each encoder to obtain more spatial pertinent features. Then, we fused the feature maps extracted from the output of each network and merged them into our decoder with an attention mechanism. The method was evaluated on the BraTS 2020 dataset to segment the different types of tumors and the results show a good performance in terms of dice similarity, with coefficients of 0.8741, 0.8069, and 0.7033 for the whole tumor, core tumor, and enhancing tumor, respectively.

A proposed architectural learner model for a personalized learning environment.

Nowadays, the need for e-learning is amplified, especially after the Covid-19 pandemic. E-learning platforms present a solution for the continuity of the learning process. Learners are using different platforms and tools for learning. For this, it is necessary to model the learner for the personalization of the learning environment according to his needs, and characteristics, which will allow having a more effective and efficient environment. The existing literature maintains that the learner model represents the basis and the key to adaptation. To achieve this goal, we propose a new adaptation aspect of the learner model by integrating relevant information such as learning style, domain-related data, assessment-related data, and affective data. It has advantages in terms of precision as it solves the problem of management uncertainty of some parameters. Our approach suggests that the combination of stereotype method, fuzzy logic, and similarity techniques is an appropriate approach for initializing and updating the learner model for learning personalization.

A Contextual Relationship Model for Deceptive Opinion Spam Detection.

The promotion of e-commerce platforms has changed the lifestyle of several people from traditional marketing to digital marketing where businesses are made online and the concurrence reached high levels. These platforms have helped the ease of purchases while providing more advantages to the customers such as benefiting from a wide range of high-quality products, low prices, buying at any time, and more importantly supplying information and reviews about the products, and so on. Unfortunately, a plethora of companies mislead the customers to buy their products or demote the competitors' by using deceptive opinion spams which has a negative impact on the decision and the behavior of the purchasers. Deceptive opinion spams are written deliberately to seem legitimate and authentic so that to misguide or delude the customer's purchases. Consequently, the detection of these opinions is a hard task due to their nature for both humans and machines. Most of the studies are based on traditional machine learning and sparse feature engineering. However, these models do not capture the semantic aspect of reviews. According to many researchers, it is the key to the detection of deceptive opinion spam. Besides, only a few studies consider using contextual information by adopting neural networks in comparison with plenty of traditional machine learning classifiers. These models face numerous shortcomings as long as their representations are obtained while mining each review considering only words, sentences, reviews, or a combination of them, thereby classifying them based on their representations. In fact, deceptive opinions are written by the same deceivers belonging to the same companies with similar aims to promote or demolish a product. In other words, Deceptive opinion spams tend to be semantically coherent with each other. To the best of our knowledge, no model tries to obtain a representation based on the contextual relationships between opinions. This article proposes to use a capsule neural network, bidirectional long short-term memory, attention mechanism, and paragraph vector distributed bag of words to detect deceptive opinion spam. Our model provides a powerful representation of the opinions since it centers on the preservation of their contexts and the relationships between them. The results show that our model significantly outperforms the existing state-of-the-art models.

Brain Tumor Segmentation Based on Deep Learning's Feature Representation.

Brain tumor is considered as one of the most serious causes of death in the world. Thus, it is very important to detect it as early as possible. In order to predict and segment the tumor, many approaches have been proposed. However, they suffer from different problems such as the necessity of the intervention of a specialist, the long required run-time and the choice of the appropriate feature extractor. To address these issues, we proposed an approach based on convolution neural network architecture aiming at predicting and segmenting simultaneously a cerebral tumor. The proposal was divided into two phases. Firstly, aiming at avoiding the use of the labeled image that implies a subject intervention of the specialist, we used a simple binary annotation that reflects the existence of the tumor or not. Secondly, the prepared image data were fed into our deep learning model in which the final classification was obtained; if the classification indicated the existence of the tumor, the brain tumor was segmented based on the feature representations generated by the convolutional neural network architectures. The proposed method was trained on the BraTS 2017 dataset with different types of gliomas. The achieved results show the performance of the proposed approach in terms of accuracy, precision, recall and Dice similarity coefficient. Our model showed an accuracy of 91% in tumor classification and a Dice similarity coefficient of 82.35% in tumor segmentation.

Content Based medical image retrieval based on BEMD: optimization of a similarity metric.

Most medical images are now digitized and stored in patients files databases. The challenge is how to use them for acquiring knowledge or/and for aid to diagnosis. In this paper, we address the challenge of diagnosis aid by Content Based Image Retrieval (CBIR). We propose to characterize images by using the Bidimensional Empirical Mode Decomposition (BEMD). Images are decomposed into a set of functions named Bidimensional Intrinsic Mode Functions (BIMF). Two methods are used to characterize BIMFs information content: the Generalized Gaussian density functions (GGD) and the Huang-Hilbert transform (HHT). In order to enhance results, we introduce a similarity metric optimization process: weighted distances between BIMFs are adapted for each image in the database. Retrieval efficiency is given for different databases (DB), including a diabetic retinopathy DB, a mammography DB and a faces DB. Results are promising: the retrieval efficiency is higher than 95% for some cases.