Udith Haputhanthri

selected publications

1. Deep Optical Coding Design in Computational Imaging

   Arguello, Henry, Bacca, Jorge, Kariyawasam, Hasindu, Vargas, Edwin, Márquez, Miguel, Hettiarachchi, Ramith, Garcia, Hans, Herath, Kithmini,  Haputhanthri, Udith, Ahluwalia, Balpreet Singh, So, Peter T. C., Wadduwage, Dushan N., and Edussooriya, Chamira U. S.

   ArXiv 2022
2. From Hours to Seconds: Towards 100x Faster Quantitative Phase Imaging via Differentiable Microscopy

   Haputhanthri, Udith, Herath, Kithmini, Hettiarachchi, Ramith, Kariyawasam, Hasindu, Ahmad, Azeem, Ahluwalia, Balpreet Singh, Edussooriya, Chamira U. S., and Wadduwage, Dushan N.

   ArXiv 2022
3. [P]

   Differentiable Microscopy Designs an All Optical Quantitative Phase Microscope

   Herath, Kithmini*,  Haputhanthri, Udith*, Hettiarachchi, Ramith*, Kariyawasam, Hasindu*, Ahmad, Azeem, Ahluwalia, Balpreet Singh, Edussooriya, Chamira U. S., and Wadduwage, Dushan N.

   ArXiv 2022

   Abs HTML PDF

   Ever since the first microscope by Zacharias Janssen in the late 16th century, scientists have been inventing new types of microscopes for various tasks. Inventing a novel architecture demands years, if not decades, worth of scientific experience and creativity. In this work, we introduce Differentiable Microscopy (∂μ), a deep learning-based design paradigm, to aid scientists design new interpretable microscope architectures. Differentiable microscopy first models a common physics-based optical system however with trainable optical elements at key locations on the optical path. Using pre-acquired data, we then train the model end-to-end for a task of interest. The learnt design proposal can then be simplified by interpreting the learnt optical elements. As a first demonstration, based on the optical 4-f system, we present an all-optical quantitative phase microscope (QPM) design that requires no computational post-reconstruction. A follow-up literature survey suggested that the learnt architecture is similar to the generalized phase concept developed two decades ago. We then incorporate the generalized phase contrast concept to simplify the learning procedure. Furthermore, this physical optical setup is miniaturized using a diffractive deep neural network (D2NN). We outperform the existing benchmark for all-optical phase-to-intensity conversion on multiple datasets, and ours is the first demonstration of its kind on D2NNs. The proposed differentiable microscopy framework supplements the creative process of designing new optical systems and would perhaps lead to unconventional but better optical designs.

4. [P]

   Differentiable Microscopy for Content and Task Aware Compressive Fluorescence Imaging

   Haputhanthri, Udith, Seeber, Andrew, and Wadduwage, Dushan N.

   ArXiv 2022

   Abs HTML PDF

   The trade-off between throughput and image quality is an inherent challenge in microscopy. To improve throughput, compressive imaging under-samples image signals; the images are then computationally reconstructed by solving a regularized inverse problem. Compared to traditional regularizers, Deep Learning based methods have achieved greater success in compression and image quality. However, the information loss in the acquisition process sets the compression bounds. Further improvement in compression, without compromising the reconstruction quality is thus a challenge. In this work, we propose differentiable compressive fluorescence microscopy (∂μ) which includes a realistic generalizable forward model with learnable-physical parameters (e.g. illumination patterns), and a novel physics-inspired inverse model. The cascaded model is end-to-end differentiable and can learn optimal compressive sampling schemes through training data. With our model, we performed thousands of numerical experiments on various compressive microscope configurations. Our results suggest that learned sampling outperforms widely used traditional compressive sampling schemes at higher compressions (\times 100 - \times 1000) in terms of reconstruction quality. We further utilize our framework for Task Aware Compression. The experimental results show superior performance on segmentation tasks even at extremely high compression (\times 4096).

5. [C] ICASSP

   Towards Accurate Cross-Domain In-Bed Human Pose Estimation

   Afham, Mohamed*,  Haputhanthri, Udith*, Pradeepkumar, Jathurshan*, Anandakumar, Mithunjha, Silva, Ashwin De, and Edussooriya, Chamira U. S.

   ArXiv 2021

   Abs HTML PDF Code

   Human behavioral monitoring during sleep is essential for various medical applications. Majority of the contactless human pose estimation algorithms are based on RGB modality, causing ineffectiveness in in-bed pose estimation due to occlusions by blankets and varying illumination conditions. Long-wavelength infrared (LWIR) modality based pose estimation algorithms overcome the aforementioned challenges; however, ground truth pose generations by a human annotator under such conditions are not feasible. A feasible solution to address this issue is to transfer the knowledge learned from images with pose labels and no occlusions, and adapt it towards real world conditions (occlusions due to blankets). In this paper, we propose a novel learning strategy comprises of two-fold data augmentation to reduce the cross-domain discrepancy and knowledge distillation to learn the distribution of unlabeled images in real world conditions. Our experiments and analysis show the effectiveness of our approach over multiple standard human pose estimation baselines.

6. [C] SPIE BiOS

   DEEP learning powered De-scattering with Excitation Patterning (DEEP) for high-throughput wide-field multiphoton microscopy

   Wijethilake, Navodini, Zheng, Cheng, Park, Jong K., Yildirim, Murat,  Haputhanthri, Udith, So, Peter T. C., and Wadduwage, Dushan N.

   In Multiphoton Microscopy in the Biomedical Sciences XXII 2022

   Abs HTML Video

   Multiphoton microscopy is the gold standard for deep tissue fluorescence imaging. Long wavelengths enable hundreds of microns deep penetration of excitation light, but the emission fluorescence at shorter wavelengths encounters scattering before detection. While not being an issue for point scanning geometries, for wide-field geometries emission light scattering degrades the image quality. In this work, we use temporally focused pattered excitations to spatially encode image information before emission light scattering. Upon detection, images are reconstructed computationally by solving a linear inverse problem. We further improve our results by learning inverse solvers and optimal patterns through physics-based deep learning.

7. [C] ISCAS

   A Novel Transfer Learning-Based Approach for Screening Pre-Existing Heart Diseases Using Synchronized ECG Signals and Heart Sounds

   Hettiarachchi, Ramith,  Haputhanthri, Udith, Herath, Kithmini, Kariyawasam, Hasindu, Munasinghe, Shehan, Wickramasinghe, Kithmin, Samarasinghe, Duminda, De Silva, Anjula, and Edussooriya, Chamira U. S.

   In 2021 IEEE International Symposium on Circuits and Systems (ISCAS) 2021

   Abs HTML PDF

   Diagnosing pre-existing heart diseases early in life is important as it helps prevent complications such as pulmonary hypertension, heart rhythm problems, blood clots, heart failure and sudden cardiac arrest. To identify such diseases, phonocardiogram (PCG) and electrocardiogram (ECG) waveforms convey important information. Therefore, effectively using these two modalities of data has the potential to improve the disease screening process. We evaluate this hypothesis on a subset of the PhysioNet Challenge 2016 Dataset which contains simultaneously acquired PCG and ECG recordings. Our novel dual-convolutional neural network based approach uses transfer learning to tackle the problem of having limited amounts of simultaneous PCG and ECG data that is publicly available, while having the potential to adapt to larger datasets. In addition, we introduce two main evaluation frameworks named record-wise and sample-wise evaluation which leads to a rich performance evaluation for the transfer learning approach. Comparisons with methods which used single or dual modality data show that our method can lead to better performance. Furthermore, our results show that individually collected ECG or PCG waveforms are able to provide transferable features which could effectively help to make use of a limited number of synchronized PCG and ECG waveforms and still achieve significant classification performance.

8. [J] IJNME

   Topologically optimal design and failure prediction using conditional generative adversarial networks

   Herath, Sumudu, and Haputhanthri, Udith

   International Journal for Numerical Methods in Engineering Dec 2021

   Abs HTML PDF Code

   Among the various structural optimization tools, topology optimization is the widely used technique in obtaining the initial design of structural components. The resulting topologically optimal initial design will be the input for subsequent structural optimizations such as shape, size and layout optimizations. However, iterative solvers used in conventional topology optimization schemes are known to be computationally expensive, thus act as a bottleneck in the manufacturing process. In this paper, a novel deep learning-based accelerated topology optimization technique with the ability to predict ductile material failure is presented. A Conditional Generative Adversarial Network (cGAN) coupled with a Convolutional Neural Network (CNN) is used to predict the optimal topology of a given structure subject to a set of input variables. Subsequently, the same cGAN is trained to predict the Von-Mises stress contours on the optimal structure by means of color transformed image-to-image translations. The ductile failure criterion is evaluated by comparing the cGAN predicted maximum Von-Mises stress with the yield strength of the material. The proposed novel numerical method is proven to arrive at the topologically optimal design, accompanying the material failure decision within a negligible amount of time but also maintaining a higher prediction accuracy.

9. [J] JACM

   Nonlinear Multiscale Modelling and Design using Gaussian Processes

   Herath, Sumudu, and Haputhanthri, Udith

   Journal of Applied and Computational Mechanics Dec 2021

   Abs HTML PDF

   A method for nonlinear material modeling and design using statistical learning is proposed to assist in the mechanical analysis of structural materials. Conventional computational homogenization schemes are proven to underperform in analyzing the complex nonlinear behavior of such microstructures with finite deformations. Also, the higher computational cost of the existing homogenization schemes inspires the inception of a data-driven multiscale computational homogenization scheme. In this paper, a statistical nonlinear homogenization scheme is discussed to mitigate these issues using the Gaussian Process Regression technique. A data-driven model is trained for different strain states of microscale unit cells. In the macroscale, nonlinear response of the macroscopic structure is analyzed, for which the stresses and material responses are predicted by the trained surrogate model.