Yazar "Hepsag, Pinar Uskaner" seçeneğine göre listele

Listeleniyor 1 - 3 / 3
  • [ X ]
    Öğe
    Predicting IVF outcomes using a logistic regression-ABC hybrid model: A proof-of-concept study on supplement associations
    (Public Library Science, 2025) Ejder, Ugur; Hepsag, Pinar Uskaner
    Machine learning models are increasingly applied to assisted reproductive technologies (ART), yet most studies rely on conventional algorithms with limited optimization. This proof-of-concept study investigates whether a hybrid Logistic Regression-Artificial Bee Colony (LR-ABC) framework can enhance predictive performance in in vitro fertilization (IVF) outcomes while producing interpretable, hypothesis-driven associations with nutritional and pharmaceutical supplement use. A retrospective dataset of 162 women undergoing IVF was analyzed. Clinical, demographic, and supplement variables were preprocessed into 21 predictors. Four algorithms (K-Nearest Neighbors, Classification and Regression Tree, Support Vector Machine, and Random Forest) were implemented alongside their LR-ABC hybrid counterparts. Model performance was evaluated using 5-fold cross-validation with Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance. Local Interpretable Model-agnostic Explanations (LIME) were applied to improve interpretability. Across all algorithm models, LR-ABC hybrids outperformed their baseline models (e.g., Random Forest: 85.2% -> 91.36% accuracy). LIME explanations identified omega-3, folic acid, and dietician support as influential features in individual predictions. However, given the small sample size, binary representation of supplements, and absence of external validation, the observed improvements and associations should be regarded as exploratory rather than definitive. The LR-ABC hybrid model demonstrates methodological potential for improving prediction and interpretability in IVF research. Findings regarding supplement associations are hypothesis-generating, not clinically directive. Future studies with larger, multi-center datasets including detailed dosage and dietary data are needed to validate and extend this framework.
  • [ X ]
    Öğe
    Transfer Learning with Fuzzy for Breast Cancer
    (Inst Information Science, 2024) Hepsag, Pinar Uskaner; Ozel, Selma Ayse; Yazici, Adnan
    Deep learning methods have been used to reduce the number of unnecessary breast biopsies. In this study, an accurate hybrid rule-based fuzzy system with transfer learning is developed to classify breast abnormalities as malignant or benign by calculating breast cancer risk from digital mammogram images. Our system consists of three phases: (i) data augmentation methods (e.g., traditional methods, Generative Adversarial Networks (GANs)); (ii) classification of the breast abnormalities on the BCDR-D02 and mini-MIAS databases by fine-tuning transfer learning methods with the deep learning base model Convolutional Neural Network (CNN); and (iii) calculation of breast cancer risk with a rule-based fuzzy system using the results of the second phase to improve the classification of breast abnormality results. Using our CNN baseline model and traditional extension methods, we achieve 64% and 82% accuracy for mini-MIAS and BCDR-D02, respectively. With fine-tuning the transfer learning methods, we obtain 80% and 83% with VGG-16 for mini- MIAS and BCDR-D02, respectively. Using the rule-based fuzzy system, called the risk method, we achieve the highest results for mini-MIAS (93%) and BCDR-D02 (94%). The classification results of our risk method are compared with the other transfer learning and baseline methods, and it is found that the accuracy of breast abnormality classification is improved by using a hybrid rule-based fuzzy system with transfer learning. Our study can serve as a guide that provides useful tips to researchers in the field of breast cancer classification to develop more effective and reliable studies.
  • [ X ]
    Öğe
    Using BERT models for breast cancer diagnosis from Turkish radiology reports
    (Springer, 2024) Hepsag, Pinar Uskaner; Ozel, Selma Ayse; Dalci, Kubilay; Yazici, Adnan
    Diagnostic radiology is concerned with obtaining images of the internal organs using radiological imaging procedures. These images are then interpreted by a diagnostic radiologist, who produces a textual report that assists in the diagnosis of illness or injury. Early detection of certain illnesses, particularly cancer, is critical, and the reports produced by diagnostic radiologists play a key role in this process. To develop models for the early detection of cancer, text classification techniques can be applied to radiological reports. However, this process requires access to a dataset of radiology reports, which is not widely available. Currently, radiology report datasets exist for high-resource languages such as English and Dutch, but not for low-resource languages such as Turkish. This article describes the collection of a mammography report dataset for Turkish, consisting of 62 reports from real patients that were manually labeled by an expert for diagnosing breast cancer. Basic machine learning models were applied to this dataset using pre-trained BERT, DistilBERT, and an ensemble learning hard voting approach. The results showed that BERT on Turkish achieved the best performance, with a 91% F1-score. Hard Voting, which combined the results of BERTTurkish, BERTClinical, and BERTMultilingual, achieved the highest F1-score of 93%. The results show that BERT and Hard Voting outperform the other machine learning models for breast cancer diagnosis from Turkish radiology reports.