Mikroskobik Görüntülerde Diyatomların Derin Öğrenme Tabanlı Segmentasyonu

Öz

Bu çalışmada, mikroskobik diyatom görüntülerinin piksel seviyesinde otomatik olarak segmentasyonu için derin öğrenme tabanlı anlamsal segmentasyon yaklaşımları ele alınmıştır. Diyatomların ince ve karmaşık morfolojik yapıları, değişken aydınlatma koşulları ve arka plan gürültüsü, klasik görüntü işleme yöntemleriyle güvenilir segmentasyon elde edilmesini zorlaştırmaktadır. Bu kapsamda, diyatom segmentasyonu problemine yönelik olarak U-Net, Attention U-Net ve DeepLabV3+ mimarileri karşılaştırmalı olarak değerlendirilmiştir. Çalışmada, 2197 adet yüksek çözünürlüklü mikroskobik görüntü ve bu görüntülere ait manuel olarak oluşturulmuş ikili segmentasyon maskelerinden oluşan bir veri seti kullanılmıştır. Tüm modeller, aynı veri bölünmesi, ortak eğitim stratejileri ve aynı değerlendirme ölçütleri altında eğitilmiştir. Segmentasyon performansı, Dice katsayısı, IoU, precision ve recall metrikleri kullanılarak analiz edilmiştir. Deneysel sonuçlar, Attention U-Net ve DeepLabV3+ modellerinin klasik U-Net’e kıyasla daha yüksek segmentasyon başarımı sunduğunu göstermiştir. Özellikle DeepLabV3+ modeli, 0,8372 Dice katsayısı ve 0,7728 IoU değeri ile en yüksek performansı elde ederek diyatom sınırlarını daha doğru ve tutarlı biçimde ayırt edebilmiştir. Görsel değerlendirmeler de, DeepLabV3+’ın ince ve uzun yapılı diyatomların segmentasyonunda daha pürüzsüz ve bütüncül sonuçlar ürettiğini ortaya koymuştur.

Anahtar Kelimeler

• Diyatom
• anlamsal segmentasyon
• derin öğrenme
• mikroskobik görüntü analizi
• konvolüsyonel sinir ağları

Deep Learning-Based Segmentation of Diatoms in Microscopic Images

Öz

This study investigates deep learning-based semantic segmentation approaches for the automatic pixel-level segmentation of microscopic diatom images. The complex morphological structures of diatoms, varying illumination conditions, and background noise make accurate segmentation challenging for traditional image processing techniques. In this study, three widely used convolutional neural network architectures—U-Net, Attention U-Net, and DeepLabV3+—were comparatively evaluated for the segmentation of diatoms in microscopic images. The experiments were conducted on a dataset consisting of 2,197 high-resolution microscopic images along with their manually annotated binary segmentation masks. All models were trained under identical experimental conditions using the same data split, training strategy, and evaluation metrics to ensure a fair comparison. Segmentation performance was assessed using Dice coefficient, Intersection over Union (IoU), precision, and recall metrics. Experimental results demonstrated that Attention U-Net and DeepLabV3+ achieved superior segmentation performance compared to the baseline U-Net model. Among the evaluated models, DeepLabV3+ produced the best results, achieving a Dice score of 0.8372 and an IoU value of 0.7728. The visual evaluation of the segmentation outputs also showed that DeepLabV3+ was able to capture thin and elongated diatom structures more accurately and consistently. These findings indicate that multi-scale contextual feature extraction plays a crucial role in improving the segmentation of complex microscopic structures such as diatoms.

Anahtar Kelimeler

• Diatom
• semantic segmentation
• deep learning
• microscopic ımage analysis
• convolutional neural networks

Kaynakça

1. Lobus NV, Kulikovskiy MS. The co-evolution aspects of the biogeochemical role of phytoplankton in aquatic ecosystems: A review. Biology 2023; 12(1): 92.
2. B-Béres V, Stenger-Kovács C, Buczkó K, Padisák J, Selmeczy GB, Lengyel E, Tapolczai K. Ecosystem services provided by freshwater and marine diatoms. Hydrobiologia 2023; 850(12): 2707-2733.
3. Barua P. Diatoms as Microbial Architects of the Silica Cycle: Ecological Significance, Biogenic Bioactives, and Emerging Biotechnological Applications. Microbial Bioactives 2025; 8(1): 1-12.
4. Costa APT, Schneck F. Diatoms as indicators in running waters: trends of studies on biological assessment and monitoring. Environmental Monitoring and Assessment 2022; 194(10): 695.
5. Hamed A. A brief and general overview on diatoms and their applications. Egyptian Journal of Phycology 2023; 24(1): 1-53.
6. Barsanti L, Birindelli L, Gualtieri P. Water monitoring by means of digital microscopy identification and classification of microalgae. Environmental Science: Processes & Impacts 2021; 23(10): 1443-1457.
7. Yu Y, Wang C, Fu Q, Kou R, Huang F, Yang B, Gao M. Techniques and challenges of image segmentation: A review. Electronics 2023; 12(5): 1199.
8. Singh S, Mittal N, Thakur D, Singh H, Oliva D, Demin A. Nature and biologically inspired image segmentation techniques. Archives of Computational Methods in Engineering 2022; 29(3): 1415-1442.

9. Mittal S, Srivastava S, Jayanth JP. A survey of deep learning techniques for underwater image classification. IEEE Transactions on Neural Networks and Learning Systems 2022; 34(10): 6968-6982.
10. Pedraza A, Bueno G, Deniz O, Cristóbal G, Blanco S, Borrego-Ramos M. Automated diatom classification (Part B): a deep learning approach. Applied Sciences 2017; 7(5): 460.
11. Kloster M, Langenkämper D, Zurowietz M, Beszteri B, Nattkemper TW. Deep learning-based diatom taxonomy on virtual slides. Scientific reports 2020; 10(1): 14416.
12. Pu S, Zhang F, Shu Y, Fu W. Microscopic image recognition of diatoms based on deep learning. Journal of phycology 2023; 59(6): 1166-1178.
13. Gündüz H, Solak CN, Günal S. Segmentation of diatoms using edge detection and deep learning. Turkish Journal of Electrical Engineering and Computer Sciences 2022; 30(6): 2268-2285.
14. Gunduz H, Gunal S. A lightweight convolutional neural network (CNN) model for diatom classification: DiatomNet. PeerJ Computer Science 2024; 10: e1970.
15. Ruiz-Santaquiteria J, Bueno G, Deniz O, Vallez N, Cristobal G. Semantic versus instance segmentation in microscopic algae detection. Engineering Applications of Artificial Intelligence 2020; 87: 103271.
16. Ronneberger O, Fischer P, Brox T. U-net: Convolutional networks for biomedical image segmentation. In: International Conference on Medical image computing and computer-assisted intervention; 5-9 October 2015; Munich, Germany: Cham: Springer international publishing. pp. 234-241.
17. Zunair H, Hamza AB. Sharp U-Net: Depthwise convolutional network for biomedical image segmentation. Computers in biology and medicine 2021; 136: 104699.
18. Ibtehaz N, Rahman MS. MultiResUNet: Rethinking the U-Net architecture for multimodal biomedical image segmentation. Neural networks 2020; 121: 74-87.
19. Iqbal A, Sharif M, Khan MA, Nisar W, Alhaisoni M. FF-UNet: a U-shaped deep convolutional neural network for multimodal biomedical image segmentation. Cognitive Computation 2022; 14(4): 1287-1302.
20. Tatli U, Budak C. Biomedical image segmentation with modified U-Net. Traitement du Signal 2023; 40(2): 523-531.
21. Gündüz, Hüseyin. 2023. Diatom Dataset. Kaggle. https://www.kaggle.com/datasets/huseyingunduz/diatom-dataset/data (Erişim Tarihi: 19.09.2025)
22. Du G, Cao X, Liang J, Chen X, Zhan Y. Medical image segmentation based on U-net: A review. Journal of Imaging Science & Technology 2020; 64(2).
23. Kazerouni IA, Dooly G, Toal D. Ghost-UNet: an asymmetric encoder-decoder architecture for semantic segmentation from scratch. IEEE Access 2021; 9: 97457-97465.
24. Saifullah S, Dreżewski R, Yudhana A, Wielgosz M, Caesarendra W. Modified U-Net with attention gate for enhanced automated brain tumor segmentation. Neural Computing and Applications 2025; 37(7): 5521-5558.
25. Li P, Zhou J, Xu X. Real-time image semantic segmentation based on improved deeplabv3+ network. Big Data and Cognitive Computing 2025; 9(6): 152.
26. Du S, Du S, Liu B, Zhang X. Incorporating DeepLabv3+ and object-based image analysis for semantic segmentation of very high resolution remote sensing images. International Journal of Digital Earth 2021; 14(3): 357-378.
27. Eelbode T, Bertels J, Berman M, Vandermeulen D, Maes F, Bisschops R, Blaschko MB. Optimization for medical image segmentation: theory and practice when evaluating with dice score or jaccard index. IEEE transactions on medical imaging 2020; 39(11): 3679-3690.
28. Hou F, Lei W, Li S, Xi J, Xu M, Luo J. Improved Mask R-CNN with distance guided intersection over union for GPR signature detection and segmentation. Automation in Construction 2021; 121: 103414.
29. Foody GM. Challenges in the real world use of classification accuracy metrics: From recall and precision to the Matthews correlation coefficient. Plos one 2023; 18(10): e0291908.
30. Cullerne Bown W. Sensitivity and specificity versus precision and recall, and related dilemmas. Journal of Classification 2024; 41(2): 402-426.