Sürdürülebilir Kalkınma Amaçları
Yayınlar
- MakaleAçık Erişim2026A Personalized Target Placement Optimization Framework for VR-Based Upper Extremity Rehabilitation
Türe, Hayati; Kalfa, Eren; Aslan, Muhammed Emin; Işık, Buket Özdemir; Topçu, Osman; Özdemir, Erhan; Sarıhan, Köksal
Featured Application: The proposed framework can be directly integrated into existing VR rehabilitation platforms to provide real-time personalized target placement for post-stroke upper extremity therapy, requiring no additional hardware beyond a standard CPU. Virtual reality (VR)-based rehabilitation is an established modality for upper extremity motor recovery; however, existing systems frequently rely on fixed, random, or therapist-tuned target placement that disregards patient-specific motor capacity and population-level priors. This study proposes a cross-patient collaborative swarm intelligence framework that derives zone-based patient profiles from real VR trajectories and augments them with a similarity-weighted cohort prior distilled from clinically similar patients’ successful trajectory clouds and zone-transition graphs. A hybrid Ant Colony Optimization (ACO)–Particle Swarm Optimization (PSO) algorithm optimizes 12 targets per session across a 27-zone ( (Formula presented.) ) workspace using a five-component fitness function encompassing reachability, zone balance, movement efficiency, heatmap-guided challenge coverage, and swarm-flow consistency. The framework was evaluated retrospectively on a single-center cohort of 36 post-stroke patients and 6373 sessions under a leakage-safe simulation protocol with 70/30 chronological splits; outcomes are model-based proxy success rates derived from each patient’s profile rather than directly observed task success. The hybrid strategy achieved a mean simulated success rate of 85.5% ± 5.5%, a 36.4% relative improvement over random placement (Wilcoxon (Formula presented.), Cohen’s (Formula presented.) ); the leakage-safe split yielded 80.1% on the held-out segment versus 61.1% for random, with no statistically significant train–test gap ( (Formula presented.) ). Ablation confirmed both PSO and ACO are individually necessary ( (Formula presented.) pp, (Formula presented.) ). Total session-start computation is 78 ms on standard CPU hardware. These findings constitute a proof-of-concept that collaborative personalized swarm optimization can substantially outperform heuristic target placement under in silico evaluation; clinical efficacy in terms of standardized motor outcome measures remains to be established in a prospective randomized controlled trial, and the findings should be replicated across centers, task modes, and a larger cohort before generalization.
- MakaleAçık Erişim2025Comparison of Classical and Deep Learning Models for Respiratory Sound Classification
Türe, Hayati; Aygün, Eren
This study aims to automatically classify normal and pathological respiratory sounds using a deep learning-based approach developed on the ICBHI 2017 respiratory sound dataset. MFCCs, Mel spectrograms, and spectral features were extracted from segmented respiratory sound recordings. These features were then used to train classical machine learning algorithms (XGBoost, SVM, KNN, Random Forest) and deep learning models (GhostNet v1–v4, EfficientNet-B0, ResNet50, MobileNetV3). The impact of data augmentation techniques was also systematically examined. The results demonstrate that the GhostNet v4 model achieved the highest performance with 89% accuracy and an F1-score of 0.89. This accuracy outperforms many existing methods reported in the literature using the same dataset. Confusion matrix analyses further indicate that the model reliably distinguishes between normal and pathological classes. These findings highlight the effectiveness of deep learning-based models in the automatic analysis of respiratory sounds and suggest promising solutions for integration into clinical decision support systems.
- BildiriYalnız künye2022Segmentation of Pectoral Muscle Region in MLO Mammography Images by Backboned U-Net
Dogan, Ramazan Ozgur; Türe, Hayati; Kayikçioglu, Temel
The pectoral muscle region on MLO mammography images appears prominently similar to suspicious areas. For this reason, Computer-Aided Detection (CAD) systems remove this region to reduce false-positive rates in the mass detection process. In some cases, the pectoral muscle region is exposed to distortions due to the superposition effects caused by the mammography technique. As a result, segmentation error rates of the pectoral muscle region, whose characteristic features are deteriorated, appear. In this study, a method to identify impaired pectoral muscle regions with MobileNetV2 backboned U-Net Deep Learning method is proposed. The proposed method was tested on 84 and 201 mammography images taken from both MIAS and InBreast databases and segmented with 1.81% and 1.92% false-negative (FN) and 0.25% and 0.37% false positive (FP) rates, respectively. Particularly for distorted pectoral muscle regions, the proposed method has been shown to outperform some pioneering studies in this area.