Yazar "Gedik, Mehmet Ali" seçeneğine göre listele

Listeleniyor 1 - 4 / 4
  • [ X ]
    Öğe
    Age Estimation from Left-Hand Radiographs with Deep Learning Methods
    (Int Information & Engineering Technology Assoc, 2021) Ozdemir, Cuneyt; Gedik, Mehmet Ali; Kaya, Yilmaz
    Bone age is estimated in pediatric medicine for medical and legal purposes. In pediatric medicine, it aids in the growth and development assessment of various diseases affecting children. In forensic medicine, it is required to determine criminal liability by age, refugee age estimation, and child-adult discrimination. In such cases, radiologists or forensic medicine specialists conduct bone age estimation from left hand-wrist radiographs using atlas methods that require time and effort. This study aims to develop a computer-based decision support system using a new modified deep learning approach to accelerate radiologists' workflow for pediatric bone age estimation from wrist radiographs. The KCRD dataset created by us was used to test the proposed method. The performance of the proposed modified IncepitonV3 model compared to IncepitonV3, MobileNetV2, EfficientNetB7 models. Acceptably high results (MAE=4.3, RMSE=5.76, and R-2=0.99) were observed with the modified IncepitonV3 transfer deep learning method.
  • [ X ]
    Öğe
    Detection of Mycobacterium tuberculosis in Ziehl-Neelsen Stained Sputum Smear Specimens Using Deep Learning Techniques
    (Wiley, 2026) Genc, Ozlem; Genc, Serpil; Ozdemir, Cuneyt; Gedik, Mehmet Ali
    The initial step in diagnosing tuberculosis involves the microscopic examination of sputum samples using acid-fast staining to identify bacilli. However, this conventional method is labor-intensive, requires specialized expertise, is susceptible to errors, and has limited sensitivity. Research literature indicates that deep learning models demonstrate significant potential for detecting acid-fast bacilli (AFB) in sputum smear preparations. This study investigates the effectiveness of deep learning methods in identifying AFB within sputum smear samples. Our objective was to assess the performance of these models in tuberculosis diagnosis based on microscopic examination and to identify improvements they could bring in terms especially of sensitivity and availability within this field. We employed several transfer learning models: DenseNet201, ResNet101V2, Xception, InceptionResNetV2, and InceptionV3. In order to determine the effectiveness of these models, basic performance metrics such as accuracy, recall, precision, and F1 score were used. Among the transfer learning models we recommended, the InceptionV3 and Xception models exhibited the highest performance, achieving 99.00% high performance across all evaluation metrics. Our findings underscore that deep learning models can be effectively utilized for rapid and accurate detection of Mycobacterium tuberculosis in acid-fast stained sputum preparations.
  • [ X ]
    Öğe
    GENDER IDENTIFICATION FROM LEFT HAND-WRIST X-RAY IMAGES WITH A HYBRID DEEP LEARNING METHOD
    (Konya Teknik Univ, 2023) Ozdemir, Cuneyt; Gedik, Mehmet Ali; Kucuker, Hudaverdi; Kaya, Yilmaz
    In forensic investigations, characteristics such as gender, age, ethnic origin, and height are important in determining biological identity. In this study, we developed a deep learning-based decision support system for gender recognition from wrist radiographs using 13,935 images collected from individuals aged between 2 and 79 years. Differences in all regions of the images, such as carpal bones, radius, ulna bones, epiphysis, cortex, and medulla, were utilized. A hybrid model was proposed for gender determination from X-ray images, in which deep metrics were combined in appropriate layers of transfer learning methods. Although gender determination from X-ray images obtained from different countries has been reported in the literature, no such study has been conducted in Turkey. It was found that gender discrimination yielded different results for males and females. Gender identification was found to be more successful in females aged between 10 and 40 years than in males. However, for age ranges of 2-10 and 40-79 years, gender discrimination was found to be more successful in males. Finally, heat maps of the regions focused on by the proposed model were obtained from the images, and it was found that the areas of focus for gender discrimination were different between males and females.
  • [ X ]
    Öğe
    Opportunistic Osteoporosis Screening from Routine Knee Radiographs Using a Deep Learning-Based Texture-Branch Cross-Attention Network
    (Springer, 2026) Kurt, Mehmet; Gedik, Mehmet Ali; Kestel, Selim; Kurt, Hasan Samet; Yilmaz, Selcuk; Ozdemir, Cuneyt
    This study systematically investigates the impact of architectural components on deep learning-based osteoporosis screening using knee radiographs from a single-center archive. A total of 580 posteroanterior and lateral knee radiographs (normal: n = 389; osteoporosis: n = 191) were included. ImageNet-pretrained backbones (ResNet50V2, VGG16, and DenseNet121) were progressively extended with texture channel packages, dual-stream (global + texture) cross-attention fusion, ordinal-aware supervised contrastive loss, and input-conditioned learnable test-time augmentation (L-TTA). All models were evaluated by fivefold stratified cross-validation using the out-of-fold area under the ROC curve (OOF AUC). TBA-Net v1 achieved the highest performance (OOF AUC = 0.8727), exceeding the strongest baseline VGG16 (0.8497) by 0.0230 AUC and ResNet50V2 (0.8101) and DenseNet121 (0.8092) by 0.0626 and 0.0635 AUC. Ablation analysis demonstrated that the texture branch, cross-attention mechanism, and contrastive loss reduced performance when introduced sequentially, but produced a synergistic effect when jointly optimized; a Shapley decomposition showed that this gain arises from the interaction between components rather than from any component alone. L-TTA Net (OOF AUC = 0.8427) generated clinically interpretable attention patterns. At the current dataset scale, performance plateaued at approximately 0.87 OOF AUC, indicating that further improvements are constrained by sampling variability rather than architectural complexity. Architectural components in small medical imaging datasets should be evaluated within a holistic optimization framework rather than in isolation, and learnable test-time augmentation weights provide interpretability beyond predictive performance. The framework is positioned as a triage aid for prioritizing densitometry referral rather than as a substitute for it.