EUVIMEDEUVIMED entdecken
Crossref

OCT Medical Image Enhancement for Improving Diagnosis Accuracy: A Wavelet-Tikhonov and Deep Learning Networks

Hayder Mustafa Mueen, Mina Zolfy Lighvan, Alireza Sokhandan, Mohammed Jabardi

Advances in Artificial Intelligence and Machine Learning · 2026 · 06 (05) · 6329-6345

DOI: 10.54364/aaiml.2026.65349 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Advances in Artificial Intelligence and Machine Learning, 2026. Verfasst von Hayder Mustafa Mueen und 3 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Retinal disease classification from OCT images is often weakened by speckle noise and poor tissue contrast, which causes existing deep networks to be prone to overfitting to imaging artifacts rather than anatomical structures. To address these problems, we proposed a decoupled framework integrating physically interpretable dual-domain image enhancement and attention-based feature extraction. To pre-process the images after DWT, Tikhonov regularization with a Laplacian operator is used to smoothen the image, dampening any speckle noise, and stabilize the latent retinal membranes (RPE, for example). This improved image output is then fed into a ResNet-50 backbone with an additional convolutional block attention module (CBAM). With the cooperation of the channel attention and spatial attention mechanisms, the model was able to learn to pay attention to pathognomonic lesions over irrelevant background structures. For OCT2017 (the standard dataset has 83484 training images and 968 independent test images), the accuracy of our proposed Wavelet-TikhonovCBAM network was 98.69% and the macro-averaged f1 score was 98.45%. Numerical results of the ablation study show that, with our proposed pipeline, PSNR and SSIM reach ∼31.57 dB and ∼0.946, respectively. Moreover, it works strongly against speckle noise and costs only < 0.3% of the arithmetic of one forward pass. The classifier has only 28.09 M parameters and 3.86 GMac. It uses less resources than a transformer-based classifier and uses Gradient-weighted Class Activation Mapping (Grad-CAM) to localize the real clinical biomarkers. The proposed architecture is a resource-efficient, transparent and accurate model useful for automated ophthalmic diagnosis.

Zitieren

APA

Mueen, H. M., Lighvan, M. Z., Sokhandan, A., & Jabardi, M. (2026). OCT Medical Image Enhancement for Improving Diagnosis Accuracy: A Wavelet-Tikhonov and Deep Learning Networks. Advances in Artificial Intelligence and Machine Learning, 06(05), 6329-6345. https://doi.org/10.54364/aaiml.2026.65349

Vancouver

Mueen HM, Lighvan MZ, Sokhandan A, Jabardi M. OCT Medical Image Enhancement for Improving Diagnosis Accuracy: A Wavelet-Tikhonov and Deep Learning Networks. Advances in Artificial Intelligence and Machine Learning. 2026;06(05):6329-6345. doi:10.54364/aaiml.2026.65349

BibTeX
@article{Mueen2026,
  author = {Mueen, Hayder Mustafa and Lighvan, Mina Zolfy and Sokhandan, Alireza and Jabardi, Mohammed},
  title = {OCT Medical Image Enhancement for Improving Diagnosis Accuracy: A Wavelet-Tikhonov and Deep Learning Networks},
  journal = {Advances in Artificial Intelligence and Machine Learning},
  year = {2026},
  volume = {06},
  number = {05},
  pages = {6329--6345},
  doi = {10.54364/aaiml.2026.65349}
}
Kostenlos starten →