ERG Verileri Üzerinde Makine Öğrenmesi ve Derin Öğrenme Yaklaşımları

Yıl 2025, Cilt: 6 Sayı: 2, 1 - 6

Adem Telli , Mehmet Ömer Kırıştıoğlu , Selim Doğanay , Gıyasettin Özcan

https://doi.org/10.53608/estudambilisim.1784360

Öz

Bu çalışmada, tabular yapıda düzenlenmiş ERG verileri üzerinde makine öğrenmesi ve derin öğrenme algoritmalarının sınıflandırma performansları incelenmiştir. Çalışmada kullanılan veri seti açık kaynak olup, 106 bireyin (74 çocuk, 32 erişkin) ölçümlerinden oluşmaktadır. Veriler, ön işleme sürecinde yeniden yapılandırılmış, öznitelikler transpoze edilerek birey bazlı tabloya dönüştürülmüş ve sınıflar “normal” ve “anormal” olarak etiketlenmiştir. Beş farklı makine öğrenmesi algoritması (SVM, Random Forest, Gradient Boosting, KNN, Logistic Regression) ve çok katmanlı yapay sinir ağı (MLP) kullanılarak karşılaştırmalı analizler yapılmıştır. Performans değerlendirmesinde doğruluk, F1-skoru, hassasiyet, duyarlılık ve ROC-AUC metrikleri dikkate alınmıştır. Sonuçlar, sınıf dengesizliğinin azınlık sınıflar üzerindeki duyarlılığı düşürdüğünü göstermiştir. Derin öğrenme modeli, veri boyutu nedeniyle sınırlı performans sergilemiş, ancak wavelet tabanlı özniteliklerin eklenmesiyle iyileşme sağlanmıştır. Ayrıca, sentetik veri üretimiyle tüm modellerde doğruluk %95’in üzerine çıkmış ve F1 skorlarında belirgin artış kaydedilmiştir. Bu bulgular, sınıf dengesizliği sorununa rağmen ERG verilerinin yapay zekâ yöntemleriyle etkin şekilde sınıflandırılabileceğini ve veri artırma stratejilerinin klinik karar destek sistemlerine katkı sağlayabileceğini ortaya koymaktadır.

Anahtar Kelimeler

Elektroretinografi , Makine Öğrenmesi , Derin Öğrenme , Sınıflandırma , Veri Dengesizliği , Sentetik Veri

Machine Learning and Deep Learning Approaches on ERG Data

Öz

: This study investigates the classification performance of machine learning and deep learning algorithms on ERG data, structured in tabular form. The dataset contains recordings from 106 individuals (74 children, 32 adults) obtained under the Photopic 2.0 protocol from open access data. During preprocessing, the data were restructured, transposed into individual-based rows, and categorized into “normal” and “abnormal” classes. Five machine learning algorithms (SVM, Random Forest, Gradient Boosting, KNN, and Logistic Regression) and a multilayer perceptron (MLP) deep learning model were applied for comparative analysis. The performance was evaluated using accuracy, F1-score, precision, recall, and ROC-AUC metrics. The results revealed that class imbalance reduced sensitivity in minority classes. The deep learning model showed limited performance due to the dataset size; however, improvements were observed when wavelet-based features were included. Moreover, synthetic data generation significantly enhanced results, with all models exceeding 95% accuracy and achieving substantial increases in F1 scores. These findings demonstrate that, despite class imbalance, ERG data can be effectively classified using artificial intelligence methods, and data augmentation strategies provide significant contributions to clinical decision-support systems.

Anahtar Kelimeler

Electroretinography , Machine Learning , Deep Learning , Classification , Data Imbalance , Synthetic Data

Kaynakça

• Riggs, L. A. 1951. Electroretinography. Physiological Reviews, 31(1), 51–92. DOI:10.1152/physrev.1951.31.1.51
• Heckenlively, J. R., Arden, G. B. 2006. Principles and Practice of Clinical Electrophysiology of Vision. MIT Press, Cambridge, 921s. ISBN:9780262083461
• Granit, R. 1933. The components of the retinal action potential in mammals and their relation to the discharge in the optic nerve. Journal of Physiology, 77(3), 207–239. DOI:10.1113/jphysiol.1933.sp002964
• Fishman, G. A., Birch, D. G., Holder, G. E., Brigell, M. G. 2001. Electroretinograms: Clinical Applications. Springer, New York, 227s. DOI:10.1007/978-1-4615-1247-9
• Frishman, L. J., Robson, J. G., Harwerth, R. S. 1996. The photopic negative response of the primate electroretinogram: Ganglion cell contributions. Investigative Ophthalmology & Visual Science, 37(2), 529–545.
• Rangaswamy, N. V., Hood, D. C., Frishman, L. J., Viswanathan, S. 2007. Photopic ERGs in primate: PhNR and ganglion cell activity. Journal of Vision, 7(6), 20. DOI:10.1167/7.6.20
• Li, B., Qiu, Z., Liu, D., He, S. 2005. Optic nerve transection-induced changes in photopic negative response and correlation with retinal ganglion cell loss. Documenta Ophthalmologica, 111(3), 193–202. DOI:10.1007/s10633-005-2629-8
• Viswanathan, S., Frishman, L. J., Robson, J. G. 2000. The photopic negative response of the flash ERG in primary open angle glaucoma. Investigative Ophthalmology & Visual Science, 41(7), 2201–2211.
• Colotto, A., Falsini, B., Salgarello, T., et al. 2000. Photopic negative response of the human ERG: loss in glaucoma. Investigative Ophthalmology & Visual Science, 41(8), 2205–2211. PMID:10892864
• Tang, J., Zhou, Y., Wang, Y. 2018. Evaluation of the photopic negative response of the ERG in optic nerve disease. Documenta Ophthalmologica, 137(1), 31–40. DOI:10.1007/s10633-018-9632-0
• Park, J. C., Chen, Y. F., Blair, N. P., Shahidi, M. 2017. Comparing three different modes of electroretinography in experimental glaucoma: Diagnostic performance and correlation to structure. Investigative Ophthalmology & Visual Science, 58(14), 5959–5965. DOI:10.1167/iovs.17-22306
• Holder, G. E. 2001. Pattern ERG and retinal ganglion cell function. Eye, 15(5), 602–607. DOI:10.1038/eye.2001.145
• Diao, T., Tanaka, Y., Ueno, S., Nishida, K. 2021. Time series classification of optic neuropathy using electroretinogram waveforms versus user-defined features. Frontiers in Medicine, 8, 771713. DOI:10.3389/fmed.2021.771713
• Li, X., Xu, Y., Zhang, H. 2020. Classification of biomedical signals using wavelet transform and machine learning. Biomedical Signal Processing and Control, 62, 102073. DOI:10.1016/j.bspc.2020.102073
• Sensors. 2023. Machine learning-based classification of ERG signals using wavelet features and deep transformer models. Sensors, 23(21), 8727. DOI:10.3390/s23218727
• Liu, T., Wang, J., Chen, X. 2019. EEG/ERG feature extraction with wavelet and deep learning. Frontiers in Neuroscience, 13, 1095. DOI:10.3389/fnins.2019.01095
• Yamashita, T., Kitaoka, Y., Mizota, A. 2022. Diagnostic performance of PhNR using handheld RETeval device. Scientific Reports, 12, 12971. DOI:10.1038/s41598-022-12971-2
• Kulyabin, M., Novikov, A., Petrova, D. 2024. Synthetic electroretinogram waveforms using GANs to improve classification. arXiv preprint, arXiv:2404.11842.
• Kulyabin, M., Ivanov, P., Petrova, D. 2025. Synthetic electroretinogram signal generation using a conditional GAN. Documenta Ophthalmologica. DOI:10.1007/s10633-025-10019-0
• Goodfellow, I., Pouget-Abadie, J., Mirza, M., et al. 2014. Generative adversarial nets. Advances in Neural Information Processing Systems (NeurIPS), 27, 2672–2680. arXiv:1406.2661
• Shorten, C., Khoshgoftaar, T. M. 2019. A survey on image data augmentation for deep learning. Journal of Big Data, 6(1), 60. DOI:10.1186/s40537-019-0197-0
• Zhdanov, A. E., Dolganov, A. Yu., Borisov, V. I., Lucian, E., Bao, X., Kazaijkin, V. N., Ponomarev, V. O., Lizunov, A. V., Ivliev, S. A. 2020. OculusGraphy: Pediatric and Adults Electroretinograms Database (Description of Research Protocols). IEEE EMBS Proceedings / Technical Report.
• Diao, T., Tanaka, Y., Ueno, S., Nishida, K. 2021. Time series classification of optic neuropathy using electroretinogram waveforms versus user-defined features. Frontiers in Medicine, 8, 771713. DOI:10.3389/fmed.2021.771713
• Yamashita, T., Kitaoka, Y., Mizota, A. 2022. Diagnostic performance of photopic negative response using handheld RETeval device. Scientific Reports, 12, 12971. DOI:10.1038/s41598-022-12971-2
• Li, X., Xu, Y., Zhang, H. 2020. Classification of biomedical signals using wavelet transform and machine learning. Biomedical Signal Processing and Control, 62, 102073. DOI:10.1016/j.bspc.2020.102073
• Diao, T., Tanaka, Y., Ueno, S., Nishida, K. 2021. Time series classification of optic neuropathy using electroretinogram waveforms versus user-defined features. Frontiers in Medicine, 8, 771713. DOI:10.3389/fmed.2021.771713
• Kulyabin, M., Petrova, D., Ivanov, P. 2023. Machine learning-based classification of ERG signals using multi-wavelet and transformer models. Sensors, 23(21), 8727. DOI:10.3390/s23218727
• Phenotyping of ABCA4 Retinopathy by Machine Learning Analysis of Full-Field Electroretinography. 2022. Investigative Ophthalmology & Visual Science, 63(7), 1251.
• Kulyabin, M., Ivanov, P., Petrova, D. 2024. Artificial intelligence for detection of retinal toxicity using ERG and mfERG signals. Scientific Reports, 14, 76943. DOI:10.1038/s41598-024-76943-4
• Yang, J., Zhang, T., Chen, W. 2023. BIOT: Cross-data biosignal learning in the wild. arXiv preprint arXiv:2305.10351.
• Zhang, X., Liu, H., & Wang, Y. 2024. Hybrid CNN–Wavelet model for biomedical time-series classification. Biomedical Signal Processing and Control, 93, 106082.