arXiv subjects

Raunak Dey

Publications and source records attributed to Raunak Dey.

8 recordsLinked to original sources

arXiv2022

Non-monotonic skewness of currents in non-equilibrium steady states

Measurements of any property of a microscopic system are bound to show significant deviations from the average, due to thermal fluctuations. For time-integrated currents such as heat, work or entropy production in a steady state, it is in fact known that there will be long stretches of fluctuations both above as well as below the average, occurring equally likely at large times. In this paper we show that for any finite-time measurement in a non-equilibrium steady state - rather counter-intuitively - fluctuations below the average are more probable. This discrepancy is higher when the system is further away from equilibrium. For overdamped diffusive processes, there is even an optimal time when time-integrated current fluctuations mostly lie below the average. We demonstrate that these effects result from the non-monotonic skewness of current fluctuations and provide evidence that they are easily observable in experiments. We also discuss their extensions to discrete space Markov jump processes and implications to biological and synthetic microscopic engines.

cond-mat.stat-mech↗

arXiv2022

Experimental verification of Arcsine laws in mesoscopic non-equilibrium and active systems

A large number of processes in the mesoscopic world occur out of equilibrium, where the time course of a system evolution becomes immensely important since it is driven principally by dissipative effects. Non-equilibrium steady states (NESS) represent a crucial category in such systems, where relaxation timescales are comparable to the operational timescales. In this study, we employ a model NESS stochastic system which comprises of a colloidal microparticle, optically trapped in a viscous fluid, externally driven by a temporally correlated noise, and show that time-integrated observables such as the entropic current, the work done on the system or the work dissipated by it, follow the three Levy arcsine laws [1], in the large time limit. We discover that cumulative distributions converge faster to arcsine distributions when it is near equilibrium and the rate of entropy production is small, because in that case the entropic current has weaker temporal autocorrelation. We study this phenomenon changing the strength of the added noise or by perturbing our system with a flow field produced by a microbubble at close proximity to the trapped particle. We confirm our experimental findings with theoretical simulations of the systems. Our work provides an interesting insight into the NESS statistics of the meso-regime, where stochastic fluctuations play a pivotal role.

cond-mat.soft↗

arXiv2021

ASC-Net: Unsupervised Medical Anomaly Segmentation Using an Adversarial-based Selective Cutting Network

In this paper we consider the problem of unsupervised anomaly segmentation in medical images, which has attracted increasing attention in recent years due to the expensive pixel-level annotations from experts and the existence of a large amount of unannotated normal and abnormal image scans. We introduce a segmentation network that utilizes adversarial learning to partition an image into two cuts, with one of them falling into a reference distribution provided by the user. This Adversarial-based Selective Cutting network (ASC-Net) bridges the two domains of cluster-based deep segmentation and adversarial-based anomaly/novelty detection algorithms. Our ASC-Net learns from normal and abnormal medical scans to segment anomalies in medical scans without any masks for supervision. We evaluate this unsupervised anomly segmentation model on three public datasets, i.e., BraTS 2019 for brain tumor segmentation, LiTS for liver lesion segmentation, and MS-SEG 2015 for brain lesion segmentation, and also on a private dataset for brain tumor segmentation. Compared to existing methods, our model demonstrates tremendous performance gains in unsupervised anomaly segmentation tasks. Although there is still room to further improve performance compared to supervised learning algorithms, the promising experimental results and interesting observations shed light on building an unsupervised learning algorithm for medical anomaly identification using user-defined knowledge.

eess.IV↗

arXiv2021

ASC-Net : Adversarial-based Selective Network for Unsupervised Anomaly Segmentation

We introduce a neural network framework, utilizing adversarial learning to partition an image into two cuts, with one cut falling into a reference distribution provided by the user. This concept tackles the task of unsupervised anomaly segmentation, which has attracted increasing attention in recent years due to their broad applications in tasks with unlabelled data. This Adversarial-based Selective Cutting network (ASC-Net) bridges the two domains of cluster-based deep learning methods and adversarial-based anomaly/novelty detection algorithms. We evaluate this unsupervised learning model on BraTS brain tumor segmentation, LiTS liver lesion segmentation, and MS-SEG2015 segmentation tasks. Compared to existing methods like the AnoGAN family, our model demonstrates tremendous performance gains in unsupervised anomaly segmentation tasks. Although there is still room to further improve performance compared to supervised learning algorithms, the promising experimental results shed light on building an unsupervised learning algorithm using user-defined knowledge.

cs.CV↗

arXiv2021

Single-shot wideband active mircorheology using multiple-sinusoid modulated Optical Tweezers

We employ multiple sinusoid modulated optical tweezers to measure the frequency dependent rheological parameters of a linear viscoelastic fluid over five decades of frequency in a single shot, hitherto not achieved using active microrheology alone. Thus, we spatially modulate a trapped probe particle embedded in a fluid medium with a combination of a square wave - which is by definition a superposition of odd sinusoidal harmonics - and a linear superposition of multiple sinusoids at a wideband frequency range, with complete control over the amplitude, frequency and relative phase of the modulating signals. For the latter, we selectively excite the particle by larger amplitudes at high frequencies where the particle response is low, thereby enabling wideband active microrheology with large signal-to-noise. This mitigates the principal issue associated with conventional active microrheology - which is low bandwidth - and also renders our method better in terms of signal to noise, and faster compared to passive microrheology. We determine the complex viscoelastic parameters of the fluid by extracting the phase response (relative to input excitation) of the probe from the experimentally recorded time series data of the probe displacement, and employing well-known theoretical correlations thereafter. We test the efficacy of our method by studying a linear viscoelastic media (polyacrylamide-water solution) at different concentrations, and find good agreement of the measured fluid parameters with known literature values.

cond-mat.soft↗

arXiv2019

Hybrid Cascaded Neural Network for Liver Lesion Segmentation

Automatic liver lesion segmentation is a challenging task while having a significant impact on assisting medical professionals in the designing of effective treatment and planning proper care. In this paper we propose a cascaded system that combines both 2D and 3D convolutional neural networks to effectively segment hepatic lesions. Our 2D network operates on a slice by slice basis to segment the liver and larger tumors, while we use a 3D network to detect small lesions that are often missed in a 2D segmentation design. We employ this algorithm on the LiTS challenge obtaining a Dice score per case of 68.1%, which performs the best among all non pre-trained models and the second best among published methods. We also perform two-fold cross-validation to reveal the over- and under-segmentation issues in the LiTS annotations.

eess.IV↗

arXiv2018

CompNet: Complementary Segmentation Network for Brain MRI Extraction

Brain extraction is a fundamental step for most brain imaging studies. In this paper, we investigate the problem of skull stripping and propose complementary segmentation networks (CompNets) to accurately extract the brain from T1-weighted MRI scans, for both normal and pathological brain images. The proposed networks are designed in the framework of encoder-decoder networks and have two pathways to learn features from both the brain tissue and its complementary part located outside of the brain. The complementary pathway extracts the features in the non-brain region and leads to a robust solution to brain extraction from MRIs with pathologies, which do not exist in our training dataset. We demonstrate the effectiveness of our networks by evaluating them on the OASIS dataset, resulting in the state of the art performance under the two-fold cross-validation setting. Moreover, the robustness of our networks is verified by testing on images with introduced pathologies and by showing its invariance to unseen brain pathologies. In addition, our complementary network design is general and can be extended to address other image segmentation problems with better generalization.

cs.CV↗

arXiv2018

Diagnostic Classification Of Lung Nodules Using 3D Neural Networks

Lung cancer is the leading cause of cancer-related death worldwide. Early diagnosis of pulmonary nodules in Computed Tomography (CT) chest scans provides an opportunity for designing effective treatment and making financial and care plans. In this paper, we consider the problem of diagnostic classification between benign and malignant lung nodules in CT images, which aims to learn a direct mapping from 3D images to class labels. To achieve this goal, four two-pathway Convolutional Neural Networks (CNN) are proposed, including a basic 3D CNN, a novel multi-output network, a 3D DenseNet, and an augmented 3D DenseNet with multi-outputs. These four networks are evaluated on the public LIDC-IDRI dataset and outperform most existing methods. In particular, the 3D multi-output DenseNet (MoDenseNet) achieves the state-of-the-art classification accuracy on the task of end-to-end lung nodule diagnosis. In addition, the networks pretrained on the LIDC-IDRI dataset can be further extended to handle smaller datasets using transfer learning. This is demonstrated on our dataset with encouraging prediction accuracy in lung nodule classification.

cs.CV↗