TY  - JOUR
AB  - Non-blind deconvolution aims to restore a sharp image from its blurred counterpart given an obtained kernel. Existing deep neural architectures are often built based on large datasets of sharp ground truth images and trained with supervision. Sharp, high quality ground truth images, however, are not always available, especially for biomedical applications. This severely hampers the applicability of current approaches in practice. In this paper, we propose a novel non-blind deconvolution method that leverages the power of deep learning and classic iterative deconvolution algorithms. Our approach combines a pre-trained network to extract deep features from the input image with iterative Richardson-Lucy deconvolution steps. Subsequently, a zero-shot optimisation process is employed to integrate the deconvolved features, resulting in a high-quality reconstructed image. By performing the preliminary reconstruction with the classic iterative deconvolution method, we can effectively utilise a smaller network to produce the final image, thus accelerating the reconstruction whilst reducing the demand for valuable computational resources. Our method demonstrates significant improvements in various real-world applications non-blind deconvolution tasks.
AU  - Chobola, T.
AU  - Müller, G.*
AU  - Dausmann, V.*
AU  - Theileis, A.*
AU  - Taucher, J.*
AU  - Huisken, J.*
AU  - Peng, T.
C1  - 69839
C2  - 55107
CY  - 10662 Los Vaqueros Circle, Po Box 3014, Los Alamitos, Ca 90720-1264 Usa
SP  - 3876-3885
TI  - Leveraging Classic Deconvolution and Feature Extraction in Zero-Shot Image Restoration.
JO  - IEEE Xplore
PB  - Ieee Computer Soc
PY  - 2023
SN  - 2375-9232
ER  - 

TY  - JOUR
AB  - Inhalation therapy using dry powder inhalers play an important role in treatment of human respiratory diseases. In targeted drug delivery, it is necessary to deliver a right amount of drug to the right place for reducing the side effects, which requires a deep understanding of the behavior of inhaled particles in the human respiratory system. The purpose of present study is to evaluate the potential of ellipsoidal particles for targeting drug delivery in a realistic model of tracheobronchial airway extends from oral cavity to the fourth generation. Ellipsoidal particles with fixed minor axis of 3.6 μm and different aspect ratio in the range of 1 to 10 are injected to the airway model at steady state flow rate using the discrete phase model in Fluent software. This simulation includes drag and gravity force acting on ellipsoidal particles. The deposition patterns of ellipsoidal particles are compared to spherical particles. The results showed that flow rate has a direct effect on particle transport and consequently on deposition pattern of both ellipsoidal and spherical particles, and most of the deposition occurs in the mouth-throat. In addition, the deposition of ellipsoidal particles in the mouth-throat region reduced as the aspect ratio increased. In conclusion, ellipsoidal particles showed more flexibility for targeting drug delivery compared to spherical particles.
AU  - Abdollahi, H.*
AU  - Nabaei, M.*
AU  - Ahookhosh, K.*
AU  - Babamiri, A.*
AU  - Farnoud, A.
C1  - 68603
C2  - 53565
CY  - 345 E 47th St, New York, Ny 10017 Usa
SP  - 28-33
TI  - Evaluating the Potential of Ellipsoidal Particles for Inhalation Therapy in Comparison to Spherical Particles.
JO  - IEEE Xplore
PB  - IEEE
PY  - 2022
SN  - 2375-9232
ER  - 

TY  - JOUR
AB  - One of the most promising approaches for unsu-pervised learning is combining deep representation learning and deep clustering. Some recent works propose to simultaneously learn representation using deep neural networks and perform clustering by defining a clustering loss on top of embedded features. However, these approaches are sensitive to imbalanced data and out-of-distribution samples. As a consequence, these methods optimize clustering by pushing data close to randomly initialized cluster centers. This is problematic when the number of instances varies largely in different classes or a cluster with few samples has less chance to be assigned a good centroid. To overcome these limitations, we introduce a new unsupervised framework for joint debiased representation learning and image clustering. We simultaneously train two deep learning models, a deep representation network that captures the data distribution, and a deep clustering network that learns embedded features and performs clustering. Specifically, the clustering network and learning representation network both take advantage of our proposed statistics pooling block that represents mean, variance, and cardinality to handle the out-of-distribution samples and class imbalance. Our experiments show that using these repre-sentations, one can considerably improve results on imbalanced image clustering across a variety of image datasets. Moreover, the learned representations generalize well when transferred to the out-of-distribution dataset.
AU  - Rezaei, M.*
AU  - Dorigatti, E.
AU  - Rügamer, D.*
AU  - Bischl, B.*
C1  - 67509
C2  - 53549
CY  - 10662 Los Vaqueros Circle, Po Box 3014, Los Alamitos, Ca 90720-1264 Usa
SP  - 55-62
TI  - Joint debiased representation learning and imbalanced data clustering.
JO  - IEEE Xplore
PB  - Ieee Computer Soc
PY  - 2022
SN  - 2375-9232
ER  -