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.

Detection of early clinical keratoconus (KCN) is a challenging task, even for expert clinicians. In this study, we propose a deep learning (DL) model to address this challenge. We first used Xception and InceptionResNetV2 DL architectures to extract features from three different corneal maps collected from 1371 eyes examined in an eye clinic in Egypt. We then fused features using Xception and InceptionResNetV2 to detect subclinical forms of KCN more accurately and robustly. We obtained an area under the receiver operating characteristic curves (AUC) of 0.99 and an accuracy range of 97–100% to distinguish normal eyes from eyes with subclinical and established KCN. We further validated the model based on an independent dataset with 213 eyes examined in Iraq and obtained AUCs of 0.91–0.92 and an accuracy range of 88–92%. The proposed model is a step toward improving the detection of clinical and subclinical forms of KCN.

Corneal diseases are the most common eye disorders. Deep learning techniques are used to perform automated diagnoses of cornea. Deep learning networks require large-scale annotated datasets, which is conceded as a weakness of deep learning. In this work, a method for synthesizing medical images using conditional generative adversarial networks (CGANs), is presented. It also illustrates how produced medical images may be utilized to enrich medical data, improve clinical decisions, and boost the performance of the conventional neural network (CNN) for medical image diagnosis. The study includes using corneal topography captured using a Pentacam device from patients with corneal diseases. The dataset contained 3448 different corneal images. Furthermore, it shows how an unbalanced dataset affects the performance of classifiers, where the data are balanced using the resampling approach. Finally, the results obtained from CNN networks trained on the balanced dataset are compared to those obtained from CNN networks trained on the imbalanced dataset. For performance, the system estimated the diagnosis accuracy, precision, and F1-score metrics. Lastly, some generated images were shown to an expert for evaluation and to see how well experts could identify the type of image and its condition. The expert recognized the image as useful for medical diagnosis and for determining the severity class according to the shape and values, by generating images based on real cases that could be used as new different stages of illness between healthy and unhealthy patients.

Researchers used different methods such as image processing and machine learning techniques in addition to medical instruments such as Placido disc, Keratoscopy, Pentacam;to help diagnosing variety of diseases that affect the eye. Our paper aims to detect one of these diseases that affect the cornea, which is Keratoconus. This is done by using image processing techniques and pattern classification methods. Pentacam is the device that is used to detect the cornea’s health; it provides four maps that can distinguish the changes on the surface of the cornea which can be used for Keratoconus detection. In this study, sixteen features were extracted from the four refractive maps along with five readings from the Pentacam software. The classifiers utilized in our study are Support Vector Machine (SVM) and Decision Trees classification accuracy was achieved 90% and 87.5%, respectively of detecting Keratoconus corneas. The features were extracted by using the Matlab (R2011 and R 2017) and Orange canvas (Pythonw).

اعتلال الشبكية الخداجي (ROP) هو السبب الأكثر شيوعًا لعمى الأطفال الذي لا رجعة فيه، ويعتمد تشخيصه وعلاجه على الدرجات الذاتية بناءً على سمات الأوعية الدموية في شبكية العين. ومع ذلك، فإن هذه الطريقة شاقة وعرضة للخطأ، لذا فإن الأساليب الآلية مرغوبة لمزيد من الدقة والإنتاجية. تهدف هذه الدراسة إلى تطوير نهج قائم على التعلم العميق للتشخيص الدقيق لمرض اعتلال الشبكية الخداجي الزائد عند الأطفال الخدج باستخدام نماذج التعلم التحويلية وتقنية تصنيف الاندماج. زودنا مركز الأمل للعيون في العيادة الخاصة في بغداد، العراق، بـ 2776 صورة لقاعدة فحص مرضى اعتلال الشبكية الخداجي بين عامي 2015 و2020، واستخدمنا هذه الصور لتدريب ثلاثة نماذج للشبكات العصبية التلافيفية العميقة (ResNet50 وDensenet161 وEfficientNetB5). تم استخدام نهج مصنف الاندماج لدمج النماذج الثلاثة من أجل تشخيص شامل ودقيق. تتميز النماذج الثلاثة بمعدلات دقة نسبية تبلغ 69.78٪ و80.57٪ و81.29٪ في تصنيفاتها الخاصة. ومع ذلك، زادت الدقة الإجمالية إلى 90.28 في المائة عند استخدام مصنف الاندماج. هذا يدل على أن الطريقة المقترحة مفيدة لتحديد اعتلال الشبكية الخداجي عند الخدج. تشير نتائج الدراسة إلى أن الطريقة المقترحة لديها...