ESIDLPD: design of an efficient exudate statistics-based incremental deep learning model to detect progression of diabetic retinopathy

Diabetic retinopathy (DR) is a leading cause of blindness in diabetic patients, emphasising the need for early detection and efficient classification techniques. Existing methodologies focus on conventional deep learning models that often struggle with incremental learning and may overlook crucial exudate-based features. These models face challenges in accurately segmenting complex features in fundus images, leading to sub-optimal classification results. We propose an innovative exudate statistics-based incremental deep learning model (ESIDLPD) to overcome these limitations. Utilising fundus images, the model accurately extracts a wide range of exudates and segments complicated features using the UNet architecture. Segmented blocks are transformed into multi-domain features and are classified into DR classes through an LSTM-based RNN process, supplemented with a VARMAx process for enhanced prediction. This approach refines categorisation, improving precision by 4.5%, accuracy by 3.5%, recall by 1.9%, AUC by 2.5%, and specificity by 1.5%, while reducing diagnostic delay by 10.4%.

From A Population to Patients: The Wisconsin Epidemiologic Study of Diabetic Retinopathy

The Wisconsin epidemiologic study of diabetic retinopathy: what can we learn at 14 years?

Rapid, bloody, and blinding diabetic retinopathy

Segmentation of the upper airway using deep learning - nnUNet.

This paper presents hourly net load forecasting, which is the forecasting of the difference between the hourly demand and the hourly power produced from the Photovoltaic (PV) system, which is the load that the utility should supply to the consumer. By comparing the forecasting of the 3 models, 1) Long Short-Term Memory (LSTM), which is a deep learning model, 2) Fully Online Sequential Extreme Learning Machine (FOS-ELM), which is an incremental learning model that does not require initial training data and 3) Online Sequential Extreme Learning Machine (OS-ELM), a model that can be incrementally learned as FOS-ELM. In addition, we proposed the initial training method for the OS-ELM model by taking the first sample obtained from working to synthesize a sufficient amount of sample for the initial training of the OS-ELM model. It was found from the experiment that in the case of fixed PV penetration rate, the LSTM model had slightly lower of error in forecasting than the other two models. In the case of increasing PV penetration rate, the FOS-ELM, and OS-ELM models, with incremental learning capacity, had significantly lower errors in forecasting than the LSTM model. When comparing only the OS-ELM model using the proposed method with the FOS-ELM model, it was found that the OS-ELM model gave lower errors in forecasting than the FOS-ELM model because it was initially trained by the synthetic sample properly,

Effect of cataract surgery on the progression of diabetic retinopathy

The Association of Alcohol Consumption with the Incidence and Progression of Diabetic Retinopathy

Autism Spectrum Disorder (ASD) is a type of developmental disorder that can have notable effects on a person’s cognitive abilities, language skills, ability to recognize objects, social interactions, and communication skills. The primary etiology of this condition is attributed to genetics, and prompt detection and intervention may mitigate the potential for the individual to face exorbitant healthcare expenses and protracted diagnostic procedures. A machine learning (ML) and deep learning architecture was developed with the capability to effectively analyze datasets of autistic toddlers, accurately classifying and identifying ASD traits. To explore the feasibility of predicting and analyzing ASD characteristics across various age cohorts, we employed multiple supervised ML models, namely support vector machine (SVM), k-nearest neighbors algorithm, and decision tree, and deep learning models, such as long short-term memory (LSTM). In this study, we analyzed the ASD screening dataset of toddlers from Saudi Arabia. The ASD screening datasets of toddlers from Kaggle were used to test these models. The first dataset includes 1054 instances and 19 toddler-related features, while the remaining datasets consist of 16 features, 507 instances, 165 normal, and 141 ASD cases. We report baseline results of behavior classification using ML and DL approaches. The SVM approach achieved 100% accuracy, whereas the LSTM approach attained 100% accuracy in terms of the accuracy metric. The developed system demonstrates the efficacy of the ASD system in detecting ASD toddlers in Saudi Arabia. Furthermore, the ASD system has the potential to assist parents in examining their children at an early stage.

The increasing number of cyber-attacks on network systems has become a significant challenge for network security under real-time network scenarios. Deep learning models have proven to be effective in identifying network attacks. However, these models require a large amount of data for training, and their implementation can be computationally expensive when deployed on large-scale networks. To overcome these issues, this paper proposes a blockchain-based deep learning model that utilizes the advantages of blockchain to enhance the efficiency and security of network attack identification and mitigation. The proposed model uses a novel Proof-of-Wireless-Trust (PoWT) consensus algorithm to validate and secure the training data, and a customized Binary Cascaded Deep Learning Model (BCDLM) for training the model w.r.t. multiple attack signatures. The blockchain-based model is designed to detect and mitigate dynamic network attacks in real-time, thereby enhancing the security of network systems. The proposed model is evaluated using different network datasets.

Lyme disease is one of the most common vector-borne infections. It typically causes cardiac illnesses, neurologic illnesses, musculoskeletal disorders, and dermatologic conditions. However, most of the time, it is poorly diagnosed due to many similarities with other diseases such as drug rash. Given the potentially serious consequences of unnecessary antimicrobial treatments, it is essential to understand frequent and uncommon diagnoses that explain symptoms in this population. Recently, deep learning models have been used for the diagnosis of various rash-related diseases. However, these models suffer from overfitting and color variation problems. To overcome these problems, an efficient stacked deep transfer learning model is proposed that can efficiently distinguish between patients infected with Lyme (+) or infected with other infections. 2nd order edge-based color constancy is used as a preprocessing approach to reduce the impact of multisource light from images acquired under different setups. The AlexNet pretrained learning model is used for building the Lyme disease diagnosis model. To prevent overfitting, data augmentation techniques are also used to augment the dataset. In addition, 5-fold cross-validation is also used. Comparative analysis indicates that the proposed model outperforms the existing models in terms of accuracy, f-measure, sensitivity, specificity, and area under the curve.

Compared to non-diabetic patients, outcome after cataract surgery was reported to be worse in diabetic patients--especially in those with diabetic retinopathy. This prospective study was planned to evaluate visual outcome, progression of diabetic retinopathy, and incidence of clinically significant macular oedema (CSME) in a homogenous group of patients with non-proliferative diabetic retinopathy (NPDR) without CSME at baseline 1 year after cataract surgery. Over a period of 18 months, all consecutive patients with mild-to-moderate diabetic retinopathy who had cataract surgery with phacoemulsification and posterior chamber lens implantation were prospectively followed up. Outcomes were assessed 1 year postoperatively and included visual acuity (VA), progression of retinopathy, and incidence of CSME. Progression of retinopathy and incidence of CSME were compared to the non-operated fellow eyes. Of 50 patients included, 42 completed the 1-year follow-up. VA improved in 85% of patients, and was better than 0,5 in 71%. Progression of retinopathy occurred in 12% of eyes after cataract surgery and in 10.8% of non-operated fellow eyes. No patient developed proliferative diabetic retinopathy in the operated eye. CSME occurred in 13 operated eyes (31%), five of them with retinal ischemia, and in five non-operated eyes (13.5%). Patients with ischemic macular oedema had the worst prognosis regarding VA. Modern cataract surgery seems to have no influence on the progression of diabetic retinopathy. A visual improvement is achieved in the majority of patients with NPDR, but poorer visual outcome is observed in patients developing macular oedema.

ABSTRACTWith the increasing reliance on the Internet of Things devices in various fields, especially in smart homes or smart factories, concerns are increasing about cyber threats targeting intrusion detection systems that protect Internet of Things (IoT) networks. Existing systems studies face a gap related to security, privacy, and data integrity, as well as a lack of recent studies that include an integrated (end‐to‐end) framework. To address this gap, a comprehensive framework was developed to secure and improve the performance of intrusion detection systems in IoT environments by combining incremental learning, hybrid encryption, and blockchain technologies in three stages. In the first stage, a model was designed to detect cyber attacks using Incremental Learning (SGD Classifier). This model has the ability to retrain itself on new data. The data is encrypted and digitally signed. A blockchain model is then developed to store the encrypted data. This constitutes the second stage. In the third stage, the incremental model is retrained on new data extracted from the blockchain over several rounds to achieve the highest model accuracy. The model is trained on the main server and then distributed, via updates, to the peripheral devices so that all devices receive the same new update. The incremental learning model was initially trained on the datasets (CIC IoT 2023 dataset). After retraining, it was tested on data the model had not seen before, namely (TON IoT2020 and CIC IDS2019). The highest accuracy of the model obtained was 99.89%. The proposed framework demonstrated its ability to secure intrusion detection systems data in IoT environments and improved its work by retraining it on new traffic data.

Granular computing, a new paradigm for solving large-scale and complex problems, has made significant progresses in knowledge discovery. Granular ball computing (GBC) is a novel granular computing method, which can rapidly generate scalable and robust information granules, that is, granular balls. However, a comprehensive index for measuring the performance of a granular ball does not exist. Furthermore, GBC lacks a mechanism to deal with dynamic decision systems. Therefore, in this study, the quality index of a granular ball is first formulated. Next, with this index, a novel granular ball rough sets model (GBRS) based on GBC is proposed. GBRS is more conducive to learning knowledge from uncertain datasets and more suited to incremental learning than the latest granular ball neighborhood rough sets model based on GBC. Subsequently, an incremental mechanism is introduced into GBRS, and two incremental learning models are developed for objects increasing in stream patterns and batch patterns, respectively. In the incremental learning process, three patterns of granular balls, that is, update, fusion, and split, were well studied when a set of objects was added to the decision system. Finally, to verify the effectiveness and efficiency, we apply GBRS and these two incremental learning models into classification tasks. Compared with four current state-of-the-art classification methods based on granular computing and four classical classifiers in machine learning, the proposed classifiers in this paper achieve higher classification accuracy as well as better efficiency on benchmark datasets.

In the field of machine olfaction, odor recognition whose focus is to accurately classify odor information and improve efficiency, is one of the important research directions. In recent years, deep learning for gas identification has been used in some researches, which significantly improves the identification accuracy. As a variant of deep learning, incremental learning is widely used in the field of image classification and can effectively solve the problem of catastrophic forgetting in image classification. Incremental learning has not been practically used in gas identification research, this paper will try to use an incremental learning method to identify and classify multiple types of gases. The main objectives of the research are as follows: 1) The experimental data set was divided into two equal parts, a data preprocessing method was used to convert the sensor array data into gas images which would be the input of incremental learning. And the pre-part was trained and classified by the incremental learning network. 2) The latter part of the dataset was taked as a newly emerging gas dataset, which would be incrementally learned and classified by incremental learning model. This study used an open source gas dataset, applied machine learning and deep learning architecture to analyze and compare with the algorithm. The proposed supervised contrastive replay (SCR) was used for data training and classification in this study, achieving a 94.63% classification accuracy, and the training time is only 69.7s.