Clinical keratoconus (KCN) detection is a challenging and time-consuming task. In the diagnosis process, ophthalmologists must revise demographic and clinical ophthalmic examinations. The latter include slit-lamb, corneal topographic maps, and Pentacam indices (PI). We propose an Ensemble of Deep Transfer Learning (EDTL) based on corneal topographic maps. We consider four pretrained networks, SqueezeNet (SqN), AlexNet (AN), ShuffleNet (SfN), and MobileNet-v2 (MN), and fine-tune them on a dataset of KCN and normal cases, each including four topographic maps. We also consider a PI classifier. Then, our EDTL method combines the output probabilities of each of the five classifiers to obtain a decision based on the fusion of probabilities. Individually, the classifier based on PI achieved 93.1% accuracy, whereas the deep classifiers reached classification accuracies over 90% only in isolated cases. Overall, the average accuracy of the deep networks over the four corneal maps ranged from 86% (SfN) to 89.9% (AN). The classifier ensemble increased the accuracy of the deep classifiers based on corneal maps to values ranging (92.2% to 93.1%) for SqN and (93.1% to 94.8%) for AN. Including in the ensemble-specific combinations of corneal maps’ classifiers and PI increased the accuracy to 98.3%. Moreover, visualization of first learner filters in the networks and Grad-CAMs confirmed that the networks had learned relevant clinical features. This study shows the potential of creating ensembles of deep classifiers fine-tuned with a transfer learning strategy as it resulted in an improved accuracy while showing learnable filters and Grad-CAMs that agree with clinical knowledge. This is a step further towards the potential clinical deployment of an improved computer-assisted diagnosis system for KCN detection to help ophthalmologists to confirm the clinical decision and to perform fast and accurate KCN treatment.
Background: Diabetes mellitus is one of the commonest chronic disorders worldwide with a rapid rise in prevalence. In Iraq its prevalence is high especially in elderly age group. Patients with type 2 diabetes mellitus have higher vulnerability for complications, whether microvascular or macrovascular. Ocular complications are common in diabetes mellitus, and comprise diabetic retinopathy, diabetic papillopathy, cataract, glaucoma, dry eye disease and diabetic keratopathy. Diabetic keratopathy involves endothelial and epithelial tissues of the cornea, leading to persistent epithelial defect, corneal erosion, or corneal ulcers.
Aim of the Study: To compare the mean corneal endothelial cell count between patients wi
... Show MoreSoftware-defined networks (SDN) have a centralized control architecture that makes them a tempting target for cyber attackers. One of the major threats is distributed denial of service (DDoS) attacks. It aims to exhaust network resources to make its services unavailable to legitimate users. DDoS attack detection based on machine learning algorithms is considered one of the most used techniques in SDN security. In this paper, four machine learning techniques (Random Forest, K-nearest neighbors, Naive Bayes, and Logistic Regression) have been tested to detect DDoS attacks. Also, a mitigation technique has been used to eliminate the attack effect on SDN. RF and KNN were selected because of their high accuracy results. Three types of ne
... Show MoreThis paper presents an IoT smart building platform with fog and cloud computing capable of performing near real-time predictive analytics in fog nodes. The researchers explained thoroughly the internet of things in smart buildings, the big data analytics, and the fog and cloud computing technologies. They then presented the smart platform, its requirements, and its components. The datasets on which the analytics will be run will be displayed. The linear regression and the support vector regression data mining techniques are presented. Those two machine learning models are implemented with the appropriate techniques, starting by cleaning and preparing the data visualization and uncovering hidden information about the behavior of
... Show MoreIt is known that images differ from texts in many aspects, such as high repetition and correlation, local structure, capacitance characteristics and frequency. As a result, traditional encryption methods can not be applied to images. In this paper we present a method for designing a simple and efficient messy system using a difference in the output sequence. To meet the requirements of image encryption, we create a new coding system for linear and nonlinear structures based on the generation of a new key based on chaotic maps.
The design uses a kind of chaotic maps including the Chebyshev 1D map, depending on the parameters, for a good random appearance. The output is a test in several measurements, including the complexity of th
... Show MoreA study is made about the size and dynamic activity of sunspot using automatically detecting Matlab code ''mySS .m'' written for this purpose which mainly finds a good estimate about Sunspot diameter (in km). Theory of the Sunspot size has been described using equations, where the growth and decay phases and the area of Sunspot could be calculated. Two types of images, namely H-alpha and HMI magnetograms, have been implemented. The results are divided into four main parts. The first part is sunspot size automatic detection by the Matlab program. The second part is numerical calculations of Sunspot growth and decay phases. The third part is the calculation of Sunspot area. The final part is to explain the Sunspot activit
... Show MoreWind farm assessment project have several steps. This paper aim on the second step which is harvesting the roads properties from the loading point to the construction point. The difficulties of examining the roads whether they are appropriate or not for transporting wind turbine components. The selected site for establishing wind farm is located at Maysan province and has 262 km distance from Um – Qaser port. The results through applying remote sensing and GIS techniques on Landsat (30 m resolution) and QuckBird (0.6 m resolution) are summarized by (2 hard turn, 18 Bridges that crossing over, and 13 Bridges passing under).
Recently the use of nanofluids represents very important materials. They are used in different branches like medicine, engineering, power, heat transfer, etc. The stability of nanofluids is an important factor to improve the performance of nanofluids with good results. In this research two types of nanoparticles, TiO2 (titanium oxide) and γ-Al2O3 (gamma aluminum oxide) were used with base fluid water. Two-step method were used to prepare the nanofluids. One concentration 0.003 vol. %, the nanoparticles were examined. Scanning Electron Microscopy (SEM), Atomic Force Microscopy (AFM) and X-ray diffraction (XRD) were used to accomplish these tests. The stability of the two types of nanofluids is measured by
... Show More