Özkan İnIk

Introduction: During implant planning in the posterior maxilla, accurate selection between internal (crestal and closed) and external (lateral window and open) sinus lifting approaches is critical to ensure surgical success and patient safety. The aim of this study was to develop and evaluate an end‐to‐end, deep learning–based clinical decision support system capable of automatically determining the necessity for internal or external sinus lifting using panoramic radiographs. Materials and Methods: This retrospective study was conducted on a dataset comprising 500 panoramic dental radiographs acquired between 2015 and 2025. The maxillary sinus, alveolar crest, and related anatomical structures were annotated by experts. A You Only Look Once Version 11 Large Segmentation (YOLOv11‐l‐seg)–based deep learning architecture was employed for the segmentation task. Segmentation outputs were further processed using an adaptive geometric analysis algorithm to automatically calculate the minimum residual alveolar bone height (RBH). Based on these measurements, the system generated surgical recommendations categorized as internal sinus lift, external sinus lift, or no sinus lifting required. Model performance was quantitatively evaluated using Mask mean average precision (mAP), recall, inference time, and frames per second (FPS) and qualitatively assessed through clinically interpretable visual outputs. Agreement between AI‐derived and expert measurements was assessed using paired samples t ‐tests, Bland–Altman analysis, and intraclass correlation coefficients (ICCs). Results: YOLOv11‐Large achieved a Mask mAP50 of 0.992, Mask mAP50–95 of 0.739, recall of 0.995, and an average inference time of 21.59 ms. AI‐derived RBH measurements showed excellent agreement with expert measurements on both sides. The ICC values were 0.955 for the right side and 0.963 for the left side. Bland–Altman analysis demonstrated minimal bias within clinically acceptable limits. The system correctly classified 147 of 160 patients, yielding an overall accuracy of 91.88%. Conclusion: The proposed deep learning–based system demonstrates clinically meaningful potential to standardize and support the surgical decision‐making process for sinus lifting technique selection using panoramic radiographs. Owing to the low radiation dose and wide availability of panoramic imaging, this approach may provide valuable support for preoperative planning in implant surgery.

The SETSCI Conference Proceedings is an Open Access publication created for the storage and indexing of proceedings | In this study, a classification problem is discussed to predict whether individuals' annual income levels are above $50,000. The UCI Adult Income Dataset, which is derived from the 1994 US Census (Census) data and is widely used in the field of machine learning, was preferred as the data source. This dataset contains 48,842 samples and 14 independent variables; It covers categorical and numerical characteristics such as age, educational status, marital status, profession. In addition, some categorical variables in the dataset have missing values and the classes of the target variable are unbalanced. Individuals with incomes below 50K account for 76%, while those above 24%. Therefore, in order to interpret model performance more accurately, metrics such as the ROC curve and AUC (Area Under Curve) were taken into account as well as the accuracy rate. In the study, after the data analysis and preprocessing process was completed, four different classification algorithms were applied: Logistic Regression, Support Vector Machines (SVM), Random Forest and Naive Bayes. The Logistic Regression model achieved an accuracy rate of 83.2%, while the SVM algorithm showed a moderate performance with an accuracy of 80.3%. While the Naive Bayes algorithm was moderately successful with an accuracy of 78.7%, the highest accuracy rate was obtained in the Random Forest algorithm with 85.1%. These results revealed that despite the unbalanced class distribution, the Random Forest algorithm is an effective method in such classification problems thanks to its high accuracy.

Convolutional neural networks (CNNs) have become very popular, as they can successfully solve problems in many areas by obtaining representations of input data at different layers with tuned hyperparameters. A CNN’s hyperparameters include design parameters (DPs), which describe the depth of the CNN and order of layers; layer parameters (LPs), which are used for each CNN layer; and training parameters, which are used for training the CNN. The performance of CNNs depends on these hyperparameters, but setting them properly remains a very difficult and important problem. Although there are studies in the literature that optimize each of these three parameter groups separately, there is a lack of methodologies for simultaneous optimization of DPs and LPs in a nested framework. This study proposes a novel method called SwarmCNN, which combines particle swarm optimization and artificial bee colony algorithms to optimize both DPs and LPs. The effectiveness of SwarmCNN was evaluated across various datasets, including Mnist, Mnist RD, Mnist BN, Mnist BI, Mnist RD + BI, Convex, Rectangles, Mnist Fashion, and CIFAR-10. The results demonstrate promising accuracy rates: 99.58%, 96.20%, 97.56%, 96.39%, 83.39%, 96.92%, 100%, 93.47%, and 84.77%, respectively. Comparative analysis against numerous competitors revealed SwarmCNN’s superiority on five datasets and its second-place ranking on four datasets. The results demonstrated that SwarmCNN emerges as a powerful and competitive solution for optimizing hyperparameters and conducting neural architecture searches with high accuracy on various datasets.

Coronavirus Disease (COVID-19) is an RNA-type virus that is spreading worldwide. COVID-19, which was first seen in Wuhan, China, in December 2019, quickly began to be seen in all countries of the world. Symptoms such as respiratory tract infections, fever, cough and shortness of breath are common in the diagnosis of the disease. The detection of the disease is done in the first stage by applying the Polymerase Chain Reaction (PCR) test. The long duration of laboratory examinations has led researchers to different methods. In this study, a model that can help radiologists detect the disease through Computed Tomography (CT) images was designed. This system, based on deep learning, aims to detect the disease by classification method through COVID-19 positive and negative chest tomography images. The data set used in the study consists of a total of 5000 images. Experimental studies have been conducted on Convolutional Neural Network (CNN) models such as AlexNet, Densenet201, GoogleNet, ResNet-50, Vgg-16, EfficientNet and the proposed CNN model. With the designed CNN model, COVID-19 was predicted with a success rate of 99.20%. An effective and successful model is proposed for COVID-19 detection from CT images.

Background/Objectives: Skin cancer is a major public health concern, where early diagnosis and effective treatment are essential for prevention. To enhance diagnostic accuracy, researchers have increasingly utilized computer vision systems, with deep learning-based approaches becoming the primary focus in recent studies. Nevertheless, there is a notable research gap in the effective optimization of hyperparameters to design optimal deep learning architectures, given the need for high accuracy and lower computational complexity. Methods: This paper puts forth a robust metaheuristic optimization-based approach to develop novel deep learning architectures for multi-class skin cancer classification. This method, designated as the SADASNet (Selective and Adaptive Deep Architecture Search Network by Hyperparameter Optimization) algorithm, is developed based on the Particle Swarm Optimization (PSO) technique. The SADASNet method is adapted to the HAM10000 dataset. Innovative data augmentation techniques are applied to overcome class imbalance issues and enhance the performance of the model. The SADASNet method has been developed to accommodate a range of image sizes, and six different original deep learning models have been produced as a result. Results: The models achieved the following highest performance metrics: 99.31% accuracy, 97.58% F1 score, 97.57% recall, 97.64% precision, and 99.59% specificity. Compared to the most advanced competitors reported in the literature, the proposed method demonstrates superior performance in terms of accuracy and computational complexity. Furthermore, it maintains a broad solution space during parameter optimization. Conclusions: With these outcomes, this method aims to enhance the classification of skin cancer and contribute to the advancement of deep learning.

Real-time multi-object detection and tracking is one of the main challenges in analysing aerial imagery from platforms such as unmanned aerial vehicles (UAVs) and satellite systems. Traditional tracking algorithms are inadequate especially in complex and dynamic environments, which necessitates the development of more powerful and flexible methods. This study proposes a deep learning based solution in line with the objectives of high extraction rate and sufficient accuracy. Within the scope of the study, the YOLOv11 model, which is a single-stage object detection architecture, is integrated with the ByteTrack algorithm for tracking the detected objects. This approach is evaluated in comparison with the two-stage Faster R-CNN architecture (MobileNetV3-large FPN backbone), which is known for its high accuracy in object detection. The training of the models was performed on a comprehensive dataset created by combining the challenging VisDrone and DOTA datasets containing objects of different sizes. The results show that the YOLOv11 model is more suitable for real-time applications due to its high speed inference capability, while the Faster R-CNN model provides more accurate detections despite its higher computational cost. The ByteTrack algorithm used in the tracking phase has increased the tracking accuracy by successfully identifying the detected objects. In this context, the advantages and limitations of both models are evaluated and it is concluded that the choice of model should be made according to the requirements of the targeted application.

Cancer, is a leading cause of disease and death worldwide, affecting both advanced industrialized and developing nations. Breast cancer, specifically among women, is a highly prevalent and serious type of cancer, making it a focal point for extensive research in the field of medicine. In the diagnosis of breast cancer, histopathological images play a crucial role because of the abundance of valuable phenotypic information they contain. To enhance the accuracy and objectivity of breast histopathological image analysis (BHIA), classification, and detection tasks are performed on these images using deep learning architecture approaches. In a preliminary experiment conducted in this paper using the Breast Cancer Histopathological Database (BreakHis), four state-of-the-art and custom CNN architectures were proposed. The experimental results demonstrate the notable performance of the proposed custom model at 40x and 200x magnification factors, reaching accuracies of 97.49% and 97.77%, surpassing other models. The ResNet-50 model achieved higher accuracy at 100x and 400x magnifications, with accuracies of 98.56% and 96.43%, respectively. Compared to other state-of-the-art models, the proposed CNN model not only shows efficient training with a significantly shorter timeframe but also features a reduced number of layers, highlighting its superior computational efficiency. Although the parameter count is higher than that of one of the models, the model strikes a favorable balance between computational efficiency and model capacity. In light of the achieved outcomes and the existing literature, forthcoming studies endeavor can be pursued further to enhance the performance values in breast cancer classification.

Since skin cancer is one of the most common types of cancer, prompt diagnosis is essential to successful treatment. Impressive performance in image-based classification tasks has been demonstrated by convolutional neural networks (CNNs), particularly in recent years. In this study, the proposed CNN model was applied to the ISIC skin cancer classification challenge. A proposed deep learning model and four popular deep CNN models (ResNet, GoogleNet, AlexNet, and VGG16) were used to classify the skin cancer images. High levels of accuracy on test data from the ISIC dataset were achieved by the proposed CNN model, according to experimental results. Preprocessing was performed on images with sizes of 64x64, 100x100, 224x224, and 128x128 pixels. The experimental results show that the proposed CNN model achieved the highest accuracy rate of 86.76% on 128x128 size images.

Brain tumors can have very dangerous and fatal effects if not diagnosed early. These are diagnosed by specialized doctors using biopsy samples taken from the brain. This process is exhausting and wastes doctors' time too much. Researchers have been working to develop a quick and accurate way for identifying and classifying brain tumors in order to overcome these drawbacks. Computer-assisted technologies are utilized to support doctors and specialists in making more efficient and accurate decisions. Deep learning-based methods are one of these technologies and have been used extensively in recent years. However, there is still a need to explore architectures with higher accuracy performance. For this purpose, in this paper proposed a novel convolutional neural network (CNN) which has twenty-four layers to multi-classify brain tumors from brain MRI images for early diagnosis. In order to demonstrate the effectiveness of the proposed model, various comparisons and tests were carried out. Three different state-of-the-art CNN models were used in the comparison: AlexNet, ShuffleNet and SqueezeNet. At the end of the training, proposed model is achieved highest accuracy of 92.82% and lowest loss of 0.2481. In addition, ShuflleNet determines the second highest accuracy at 90.17%. AlexNet has the lowest accuracy at 80.5% with 0.4679 of loss. These results demonstrate that the proposed CNN model provides greater precision and accuracy than the state-of-art CNN models.

Automatic classification of aerial images has become one of the topics studied in recent years. Especially for the use of drones in different fields such as agricultural applications, smart city applications, surveillance and security applications, it is necessary to automatically classify the images obtained with the camera during autonomous mission execution. For this purpose, researchers have created new data sets and some computer vision methods have been developed to achieve high accuracy. However, in addition to increasing the accuracy of the developed methods, the computational complexity should also be reduced. Because the methods to be used in devices such as drones where energy consumption is important should have low computational complexity. In this study, firstly, five different state-of-art deep learning models were used to obtain high accuracy values in the classification of aerial images. Among these models, the VGG19 model achieved the highest accuracy with 94.21%. In the second part of the study, the parameters of this model were analyzed and the model was reconstructed. The number of 143.6 million parameters of the VGG19 model was reduced to 34 million. The accuracy of the model obtained by reducing the number of parameters is 93.56% on the same test data. Thus, despite the 66.5% decrease in the parameter ratio, there was only a 0.7% decrease in the accuracy value. When compared to previous studies, the results show improved performance.