iSeg-WNet: Bebek Beyin MRI Görüntülerinin Volumetrik Bölütlemesi

Bebek beyin gelişiminin incelenmesi, doğabilecek beyin fonksiyon bozuklarının erken teşhisi açısından son derece önemlidir. Beyin MRI’ larının, beyaz madde (WM), gri madde (GM) ve beyin omurilik sıvısı (CSF) dokularının bölütleme işlemi ile incelenmektedir. Bebek beyinlerinde dokular arasındaki düşük yoğunluklu kontrast bölütleme işlemini zorlaştırmaktadır. Son dönemlerde geliştirilen Derin Öğrenme mimarileri ile bölütleme işleminin son derece çok iyi yapıldığı görülmektedir. Bu çalışmada bebek beyin MRI görüntülerinin bölütlenmesi için Derin Öğrenme tabanlı iSeg-WNet adıyla bir mimari önerilmiştir. Farklı çalışmalar ile uygun hiperparametreler belirlenmiş ve farklı mimarilerin performansları karşılaştırılmıştır. Performans karşılaştırılması Dice metriğine göre yapılmıştır. Yapılan deneysel çalışmalarda, T1w ve T2w çekimlerdeki MRI görüntülerinin beraber kullanılması bölütleme performansının artırdığı gözlemlenmiştir. Aynı zamanda maliyet fonksiyonu olarak Dice Loss ve veri normalizasyon işlemi olarak da MinMax normalizasyonun kullanılması ile yüksek başarım elde edilmiştir. Farklı mimarilerin bölütleme performansları incelendiğinde, önerilen mimari ile CSF, GM ve WM dokularını en yüksek başarı ile bölütlediği görülmüştür. Önerilen mimariye https://github.com/GaffariCelik/iSeg-WNet adresinden erişilebilir.

Anahtar Kelimeler:

Derin Öğrenme, CNN, Bölütleme, Dice Loss, 3D MRI Bölütleme, iseg-2019, iseg-2017

iSeg-WNet: Volumetric Segmentation of Infant Brain MRI Images

Examination of infant brain development is extremely important in terms of early diagnosis of possible brain dysfunctions. Brain MRIs are examined by segmentation of white matter (WM), gray matter (GM) and cerebrospinal fluid (CSF) tissues. Low-density contrast between tissues in infant brains complicates the segmentation process. It is seen that the segmentation process is done very well with the Deep Learning architectures that have been developed recently. In this study, an architecture called Deep Learning-based iSeg-WNet is proposed for segmentation of infant brain MRI images. Appropriate hyperparameters were determined by different studies and the performances of different architectures were compared. Performance comparison was made according to Dice metric. In experimental studies, it has been observed that the use of MRI images in T1w and T2w images together increases the segmentation performance. At the same time, high performance was obtained by using Dice Loss as a cost function and MinMax normalization as a data normalization process. When the segmentation performances of different architectures are examined, it is seen that the proposed architecture segments CSF, GM and WM textures with the highest success. The proposed architecture is available at https://github.com/GaffariCelik/iSeg-WNet.

Keywords:

Deep Learning, CNN, Segmentation, Dice Loss, 3D MRI Segmentation, iseg-2019, iseg-2017,

PDF

___

[1] Ghosal ,P. Chowdhury, Kumar, T. A. Bhadra, A. K. Chakraborty, J. Nandi, D. 2021. MhURI:A Supervised Segmentation Approach to Leverage Salient Brain Tissues in Magnetic Resonance Images. Comput. Methods Programs Biomed., 200, 105841, doi: 10.1016/j.cmpb.2020.105841.
[2] Balafar, M. A. Ramli,A. R. Saripan, M. I. Mashohor, S. 2010. Review of brain MRI image segmentation methods. Artif. Intell. Rev., 33(3), 261–274, 2010, doi: 10.1007/s10462-010-9155-0.
[3] Jenkinson, M. Beckmann, C. F. Behrens, T. E. J. Woolrich, M. W. Smith, S. M. 2012. FSL, NeuroImage, 62, 782–790, doi: 10.1016/j.neuroimage.2011.09.015.
[4] Dai,Y. Shi, F. Wang,L. Wu, G. Shen, D. 2013. IBEAT: A toolbox for infant brain magnetic resonance image processing. Neuroinformatics, 11(2), 211–225, doi: 10.1007/s12021-012-9164-z.
[5] Fischl, B. 2012. FreeSurfer. Neuroimage, 62(2), 774–781, doi: 10.1016/j.neuroimage.2012.01.021.
[6] Mostapha, M. Styner, M. 2019. Role of deep learning in infant brain MRI analysis. Magn. Reson. Imaging, 64(June), 171–189, doi: 10.1016/j.mri.2019.06.009.
[7] Wang, L. et al. 2014. Segmentation of neonatal brain MR images using patch-driven level sets. Neuroimage, 84, 141–158, doi: 10.1016/j.neuroimage.2013.08.008.
[8] Dolz, J. Desrosiers, C. Wang, L. Yuan, J. Shen, D. Ayed, I. B. 2020. Deep CNN ensembles and suggestive annotations for infant brain MRI segmentation. Comput. Med. Imaging Graph., 79, 101660, doi: 10.1016/j.compmedimag.2019.101660.
[9] Wang, L. et al. 2019. “Benchmark on automatic six-month-old infant brain segmentation algorithms: The iSeg-2017 challenge. IEEE Trans. Med. Imaging, 38(9), 2219–2230, doi: 10.1109/TMI.2019.2901712.
[10] Bui, T. D. Shin, J. Moon, T. 2019. Skip-connected 3D DenseNet for volumetric infant brain MRI segmentation. Biomed. Signal Process. Control, 54, 101613, 2019, doi: 10.1016/j.bspc.2019.101613.
[11] Çelik, G. Talu, M. F. 2020. Resizing and cleaning of histopathological images using generative adversarial networks. Phys. A Stat. Mech. its Appl., 554, 122652, doi: 10.1016/j.physa.2019.122652.
[12] Çelik, G. Talu, M. F. 2022. A new 3D MRI segmentation method based on Generative Adversarial Network and Atrous Convolution. Biomed. Signal Process. Control, 71(PA), 103155, doi: 10.1016/j.bspc.2021.103155.
[13] Akkus, Z. Galimzianova, A. Hoogi, A. Rubin, D. L. Erickson, B. J. 2017. Deep Learning for Brain MRI Segmentation: State of the Art and Future Directions. J. Digit. Imaging, 30(4), 449–459, doi: 10.1007/s10278-017-9983-4.
[14] Yang, X. Kwitt, R. Styner, M. Niethammer, M. 2017. Quicksilver: Fast predictive image registration – A deep learning approach. Neuroimage, 158(July), 378–396, doi: 10.1016/j.neuroimage.2017.07.008.
[15] Çelik, G. Talu, M. F. 2021. Generating the image viewed from EEG signals. Pamukkale Univ. J. Eng. Sci., 27(2), 129–138, doi: 10.5505/pajes.2020.76399.
[16] Kooi, T. et al. 2017. Large scale deep learning for computer aided detection of mammographic lesions. Med. Image Anal., 35, 303–312, doi: 10.1016/j.media.2016.07.007.
[17] Souza, J. C. Bandeira Diniz, J. O. Ferreira, J. L. França da Silva, G. L. Corrêa Silva, A. de Paiva, A. C. 2019. An automatic method for lung segmentation and reconstruction in chest X-ray using deep neural networks. Comput. Methods Programs Biomed., 177, 285–296, 2019, doi: 10.1016/j.cmpb.2019.06.005.
[18] Gaál, G. Maga, B. Lukács, A. 2020. Attention U-net based adversarial architectures for chest X-ray lung segmentation. CEUR Workshop Proc., 2692, 1–7.
[19] Başaran, E. 2022. Classification of white blood cells with SVM by selecting SqueezeNet and LIME properties by mRMR method. Signal, Image Video Process., doi: 10.1007/s11760-022-02141-2.
[20] Talo, M. Yildirim, O. Baloglu, U. B. Aydin, G. Acharya, U. R. 2019. Convolutional neural networks for multi-class brain disease detection using MRI images. Comput. Med. Imaging Graph., 78, 101673, doi: 10.1016/j.compmedimag.2019.101673.
[21] Yıldırım, Ö. Pławiak, P. Tan, R. S. Acharya, U. R. 2018. Arrhythmia detection using deep convolutional neural network with long duration ECG signals. Comput. Biol. Med., 102(September), 411–420, doi: 10.1016/j.compbiomed.2018.09.009.
[22] Hannun A. Y. et al. 2019. Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network. Nat. Med., 25(1), 65–69, 2019, doi: 10.1038/s41591-018-0268-3.
[23] Acharya, U. R. et al., 2017. A deep convolutional neural network model to classify heartbeats. Comput. Biol. Med., 89(August), 389–396, doi: 10.1016/j.compbiomed.2017.08.022.
[24] Rajpurkar , P. et al. 2017. CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning. arxiv, 3–9, http://arxiv.org/abs/1711.05225.
[25] ÇALIŞAN , M. TALU, M. F. 2020. Comparison of Methods for Determining Activity from Physical Movements. J. Polytech., 0900(1), 17–23, doi: 10.2339/politeknik.632070.
[26] Özcan, T. 2020. Yığınlanmış Özdevinimli Kodlayıcılar ile Göğüs Kanserinin Sınıflandırılması ve Klasik Makine Öğrenme Metotları ile Performans Karşılaştırması. Erciyes Univ. J. Institue Sci. Technol., 36(2), 2020, https://dergipark.org.tr/tr/pub/erciyesfen/726739.
[27] Bozdag, Z. Talu,F. M. 2021. Pyramidal nonlocal network for histopathological image of breast lymph node segmentation. Int. J. Comput. Intell. Syst., 14(1), 122–131, doi: 10.2991/ijcis.d.201030.001.
[28] Sun , Y. et al. 2021. Multi-Site Infant Brain Segmentation Algorithms: The iSeg-2019 Challenge. IEEE Trans. Med. Imaging, 40(5), 1363–1376, 2021, doi: 10.1109/TMI.2021.3055428.
[29] Subramanian, N. Elharrouss, O. Al-Maadeed, S. Chowdhury, M. 2022. A review of deep learning-based detection methods for COVID-19. Comput. Biol. Med., 143, 105233,doi: 10.1016/j.compbiomed.2022.105233.
[30] Valizadeh , M. Wolff, S. J. 2022. Convolutional Neural Network applications in additive manufacturing : A review. Adv. Ind. Manuf. Eng., 4, 100072, doi: 10.1016/j.aime.2022.100072.
[31] Lecun, Y. Bengio, Y. Hinton, G. 2015. Deep learning. Nature, 521(7553), 436–444, doi: 10.1038/nature14539.
[32] Ulyanov, D. Vedaldi, A. Lempitsky, V. 2016. Instance Normalization: The Missing Ingredient for Fast Stylizationç arxiv, 2016, http://arxiv.org/abs/1607.08022.
[33] Cirillo, M. D. Abramian, D. Eklund, A. 2020. Vox2Vox: 3D-GAN for Brain Tumour Segmentation. arXiv, 1–10.
[34] Çiçek, Ö. Abdulkadir, A. Lienkamp, S. S. Brox, T. Ronneberger, O. 2016. 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation. Medical Image Computing and Computer-Assisted Intervention, 424–432.