Obando, Marcos A. (2022) Tomografía de proyección óptica acelerada mediante aprendizaje profundo / Deep learning methods for accelerated optical projection tomography. Maestría en Ingeniería, Universidad Nacional de Cuyo, Instituto Balseiro.
![[img]](/style/images/fileicons/application_pdf.png)
| PDF (Tesis) Español 2984Kb |
Resumen en español
Recientemente, el campo de los problemas lineales inversos se ha beneficiado enormemente de técnicas basadas en aprendizaje profundo en la tarea esencial de recuperar una señal a partir de un bajo número de muestras. En el campo de la reconstrucción de imágenes, los recientes abordajes involucran la extensión de la resolución de problemas inversos mediante redes neuronales profundas como funciones promotoras a dominios ralos. Por su ventaja computacional, la evaluación de estos entornos, conocidos como redes profundas desenrolladas, usualmente se halla limitada a la reconstrucción de imágenes por resonancia magnética (MRI). En este trabajo, proponemos ToMoDL (Tomographic Model-based Deep Learning), una arquitectura desenrollada para tratar el problema de acelerar la adquisición de tomografías de proyección óptica (OPT), una técnica mesoscópica para obtener imágenes de muestras biológicas translúcidas. Utilizando doce volúmenes de proyecciones angulares de peces cebra (Danio Rerio) conteniendo diferentes secciones y días post fertilización de 5 especímenes, nuestro enfoque aborda un conjunto de datos con alta variabilidad en términos de intensidad y estructura, así como un problema usualmente pasado por alto en la aplicación de redes basadas en modelo: tomografía computada [3]. Dado que en su mayoría, los dispositivos OPT son diseñados a medida del laboratorio que hace su uso, las proyecciones crudas han sido tratadas cuidadosamente para evitar artefactos de reconstrucción producidos por desalineación del sistema óptico. Por otro lado, la integración del operador de Radon en un esquema de diferenciación automática ha sido problemática hasta la aparición de recientes avances en soluciones compatibles con librerías de aprendizaje profundo. Mediante la minimización del error cuadrático medio y la medida del índice de similaridad estructural en un mapeo desenrollado con un número finito de iteraciones, la validación cruzada de ToMoDL muestra resultados de alta calidad incluso con un 5% de las proyecciones adquiridas. Los mismos usualmente superan el desempeño de reconstrucción que el estado del arte, conformado por métodos de compressed sensing, resuelto mediante Two-Step Iterative Shrinkage/Thresholding (TwIST) y arquitecturas supervisadas de tipo U-Net. Por ultimo, desarrollamos un análisis extensivo de ciertos inconvenientes en la reconstrucción anatómica de la estructura del pez cebra, para los cuales las soluciones analíticas presentan la mejor alternativa.
Resumen en inglés
Linear inverse problems have greatly benefited from deep learning techniques in the paramount goal of recovering a signal from a small number of measurements. In the field of image reconstruction, recent approaches involve the augmentation of traditional inverse problem solvers with neural networks as sparsifying functions. Given its computational tractability, gold standards for validating these frameworks, often known as deep unrolled architectures, are usually limited to magnetic resonance imaging (MRI) reconstructions. We propose ToMoDL (Tomographic Model-based Deep Learning), a deep unrolled architecture for tackling the problem of accelerating the acquisition of optical projection tomography (OPT), a mesoscopic technique for imaging biological translucid samples. Using twelve volumes of zebrafish (Danio Rerio) angular projections from four longitudinal sections and different days post-fertilisation, our approach deals with an extremely variable dataset in terms of intensity and structure, as well as an often overlooked problem in model-based deep learning: tomography reconstruction. Since many, if not most, OPT devices are custom-built, raw projections have been carefully curated to avoid reconstruction artifacts due to misalignment. On the other hand, integrating Radon operator blocks into an automatic differentiation scheme has been thorny until recent advances in PyTorch-compatible solutions arose. By minimizing the mean square error and structural similarity index metric in a fixed-iteration unrolled mapping, our cross-validated results show a reasonably high quality reconstruction with even a 5% of the acquired projections, achieving a considerably better performance than compressed sensing methods such as Two-Step Iterative Shrinkage/Thresholding (TwIST) and U-Net architectures. We also analyze problematic reconstruction issues regarding the anatomical structure of zebrafish, where analytical solutions may be chosen instead.
| Tipo de objeto: | Tesis (Maestría en Ingeniería) |
|---|---|
| Palabras Clave: | [Optical tomography; Tomografía óptica; Accelerated acquisition; Adquisición acelerada; Deep learning; Aprendizaje profundo] |
| Referencias: | [1] Davis, S. P., Kumar, S., Alexandrov, Y., Bhargava, A., da Silva Xavier, G., Rutter, G. A., et al. Convolutional neural networks for reconstruction of undersampled optical projection tomography data applied to in vivo imaging of zebrafish. Journal of biophotonics, 12 (12), e201900128, 2019. x, 17, 18 [2] Ronneberger, O., Fischer, P., Brox, T. U-net: Convolutional networks for biomedical image segmentation. En: International Conference on Medical image computing and computer-assisted intervention, p´ags. 234–241. Springer, 2015. x, 17, 18 [3] Gilton, D., Ongie, G., Willett, R. Deep equilibrium architectures for inverse problems in imaging. IEEE Transactions on Computational Imaging, 7, 1123–1133, 2021. xv, 3 [4] Kak, A. C., Slaney, M. Principles of computerized tomographic imaging. SIAM, 2001. 1, 8 [5] Radon, J. On the determination of functions from their integral values along certain manifolds. IEEE transactions on medical imaging, 5 (4), 170–176, 1986. 1 [6] Bhid´e, A., Datar, S., Stebbins, K. Case histories of significant medical advances, 2019. 1 [7] Stewart, R. R. Exploration seismic tomography: Fundamentals. Society of Exploration Geophysicists, 1991. 1 [8] Sharpe, J., Ahlgren, U., Perry, P., Hill, B., Ross, A., Hecksher-Sørensen, J., et al. Optical projection tomography as a tool for 3d microscopy and gene expression studies. Science, 296 (5567), 541–545, 2002. 1, 5 [9] Correia, T., Lockwood, N., Kumar, S., Yin, J., Ramel, M.-C., Andrews, N., et al. Accelerated optical projection tomography applied to in vivo imaging of zebrafish. PLOS one, 10 (8), e0136213, 2015. 1, 32 [10] Donoho, D. L. Compressed sensing. IEEE Transactions on information theory, 52 (4), 1289–1306, 2006. 1, 14 [11] Ongie, G., Jalal, A., Metzler, C. A., Baraniuk, R. G., Dimakis, A. G., Willett, R. Deep learning techniques for inverse problems in imaging. IEEE Journal on Selected Areas in Information Theory, 1 (1), 39–56, 2020. 2, 18, 20, 24, 32, 43 [12] Mur, A. L., Bataille, P., Peyrin, F., Ducros, N. Deep expectation-maximization for image reconstruction from under-sampled poisson data. En: 2021 IEEE 18th International Symposium on Biomedical Imaging (ISBI), págs. 1535–1539. IEEE, 2021. 2 [13] Watson, T., Andrews, N., Davis, S., Bugeon, L., Dallman, M. D., McGinty, J. Optim: Optical projection tomography integrated microscope using open-source hardware and software. PLoS One, 12 (7), e0180309, 2017. 2 [14] Aggarwal, H. K., Mani, M. P., Jacob, M. Modl: Model-based deep learning architecture for inverse problems. IEEE transactions on medical imaging, 38 (2), 394–405, 2018. 3, 18, 24, 28, 32 [15] Huang, Z., Ye, S., McCann, M. T., Ravishankar, S. Model-based reconstruction with learning: from unsupervised to supervised and beyond. arXiv preprint ar-Xiv:2103.14528, 2021. 3 [16] Szczykutowicz, T. P., Toia, G. V., Dhanantwari, A., Nett, B. A review of deep learning ct reconstruction: Concepts, limitations, and promise in clinical practice. Current Radiology Reports, págs. 1–15, 2022. 3 [17] Huisken, J., Swoger, J., Del Bene, F., Wittbrodt, J., Stelzer, E. H. Optical sectioning deep inside live embryos by selective plane illumination microscopy. Science, 305 (5686), 1007–1009, 2004. 5 [18] Bassi, A., Schmid, B., Huisken, J. Optical tomography complements light sheet microscopy for in toto imaging of zebrafish development. Development, 142 (5), 1016–1020, 2015. 5 [19] Vallejo Ramirez, P. Optical imaging methods for the study of disease models from the nano to the mesoscale. Tesis Doctoral, University of Cambridge, 2021. 7, 39 [20] Vallejo Ramirez, P. P., Zammit, J., Vanderpoorten, O., Riche, F., Bl´e, F.-X., Zhou, X.-H., et al. Optij: Open-source optical projection tomography of large organ samples. Scientific reports, 9 (1), 1–9, 2019. 10, 39 [21] Walls, J. R., Sled, J. G., Sharpe, J., Henkelman, R. M. Correction of artefacts in optical projection tomography. Physics in Medicine & Biology, 50 (19), 4645, 2005. 10 [22] Tarantola, A. Popper, bayes and the inverse problem. Nature physics, 2 (8), 492–494, 2006. 13 [23] Hadamard, J. Sur les probl`emes aux d´eriv´ees partielles et leur signification physique. Princeton university bulletin, págs. 49–52, 1902. 13 [24] Chambolle, A. An algorithm for total variation minimization and applications. Journal of Mathematical imaging and vision, 20 (1), 89–97, 2004. 13, 15 [25] Chambolle, A., De Vore, R. A., Lee, N.-Y., Lucier, B. J. Nonlinear wavelet image processing: variational problems, compression, and noise removal through wavelet shrinkage. IEEE Transactions on Image Processing, 7 (3), 319–335, 1998. 13 [26] Tremoulheac, B. R. Low-rank and sparse reconstruction in dynamic magnetic resonance imaging via proximal splitting methods. Tesis Doctoral, UCL (University College London), 2015. 14, 15 [27] Bioucas-Dias, J. M., Figueiredo, M. A. A new twist: Two-step iterative shrinkage/ thresholding algorithms for image restoration. IEEE Transactions on Image processing, 16 (12), 2992–3004, 2007. 14, 32 [28] LeCun, Y., Bengio, Y., Hinton, G. Deep learning. nature, 521 (7553), 436–444, 2015. 15, 17 [29] Rumelhart, D. E., Hinton, G. E., Williams, R. J. Learning representations by back-propagating errors. nature, 323 (6088), 533–536, 1986. 15 [30] Denker, J., Gardner, W., Graf, H., Henderson, D., Howard, R., Hubbard, W., et al. Neural network recognizer for hand-written zip code digits. Advances in neural information processing systems, 1, 1988. 16 [31] LeCun, Y., Boser, B., Denker, J. S., Henderson, D., Howard, R. E., Hubbard, W., et al. Backpropagation applied to handwritten zip code recognition. Neural computation, 1 (4), 541–551, 1989. 16 [32] Daylight, E. G. The advent of recursion in programming, 1950s-1960s, 2010. 17 [33] Bai, S., Kolter, J. Z., Koltun, V. Trellis networks for sequence modeling. arXiv preprint arXiv:1810.06682, 2018. 18 [34] Bai, S., Kolter, J. Z., Koltun, V. Deep equilibrium models. Advances in Neural Information Processing Systems, 32, 2019. 18, 43 [35] Hornik, K., Stinchcombe, M., White, H. Multilayer feedforward networks are universal approximators. Neural networks, 2 (5), 359–366, 1989. 18 [36] Zhang, K., Zuo, W., Chen, Y., Meng, D., Zhang, L. Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising. IEEE transactions on image processing, 26 (7), 3142–3155, 2017. 24 [37] Kolarik, M., Burget, R., Riha, K. Comparing normalization methods for limited batch size segmentation neural networks. En: 2020 43rd International Conference on Telecommunications and Signal Processing (TSP), págs. 677–680. IEEE, 2020. 25 [38] Ronchetti, M. Torchradon: Fast differentiable routines for computed tomography. arXiv preprint arXiv:2009.14788, 2020. 25 [39] Hestenes, M. R., Stiefel, E. Methods of conjugate gradients for solving. Journal of research of the National Bureau of Standards, 49 (6), 409, 1952. 25 [40] Kingma, D. P., Ba, J. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014. 28 [41] Pascanu, R., Mikolov, T., Bengio, Y. On the difficulty of training recurrent neural networks. En: International conference on machine learning, págs. 1310–1318. PMLR, 2013. 28 [42] Zhu, B., Liu, J. Z., Cauley, S. F., Rosen, B. R., Rosen, M. S. Image reconstruction by domain-transform manifold learning. Nature, 555 (7697), 487–492, 2018. 31 [43] Biswas, S., Aggarwal, H. K., Jacob, M. Dynamic mri using model-based deep learning and storm priors: Modl-storm. Magnetic resonance in medicine, 82 (1), 485–494, 2019. 41 [44] Zhu, X., Cheng, D., Zhang, Z., Lin, S., Dai, J. An empirical study of spatial attention mechanisms in deep networks. En: Proceedings of the IEEE/CVF international conference on computer vision, págs. 6688–6697. 2019. 43 [45] Wang, Y., Song, X., Chen, K. Channel and space attention neural network for image denoising. IEEE Signal Processing Letters, 28, 424–428, 2021. 43 [46] Sanders, T., Dwyer, C. Subsampling and inpainting approaches for electron tomography. Ultramicroscopy, 182, 292–302, 2017. 44 [47] Hendriksen, A. A., Pelt, D. M., Batenburg, K. J. Noise2inverse: Self-supervised deep convolutional denoising for tomography. IEEE Transactions on Computational Imaging, 6, 1320–1335, 2020. 44 [48] Hendriksen, A. A., Buhrer, M., Leone, L., Merlini, M., Vigano, N., Pelt, D. M., et al. Deep denoising for multi-dimensional synchrotron x-ray tomography without high-quality reference data. Scientific reports, 11 (1), 1–13, 2021. 44 [49] Antun, V., Renna, F., Poon, C., Adcock, B., Hansen, A. C. On instabilities of deep learning in image reconstruction and the potential costs of ai. Proceedings of the National Academy of Sciences, 117 (48), 30088–30095, 2020. 44 [50] Feng, C.-M., Yan, Y., Wang, S., Xu, Y., Shao, L., Fu, H. Specificity-preserving federated learning for mr image reconstruction. IEEE Transactions on Medical Imaging, 2022. 44 |
| Materias: | Medicina > Procesamiento de imágenes |
| Divisiones: | Gcia. de área de Investigación y aplicaciones no nucleares > Gcia. de Física > Física médica |
| Código ID: | 1146 |
| Depositado Por: | Tamara Cárcamo |
| Depositado En: | 11 Aug 2023 11:48 |
| Última Modificación: | 11 Aug 2023 11:48 |
Personal del repositorio solamente: página de control del documento
