arXiv subjects

Khayrul Islam

Publications and source records attributed to Khayrul Islam.

8 recordsLinked to original sources

arXiv2026

Variable-Granularity Tokenization for High-Resolution Object Detection

ViT detectors fix a uniform token grid before any learned stage. A native-resolution aerial detector must then choose between resolving few-pixel objects and staying inside compute and memory limits. We introduce VGTok, a training-free tokenizer that sets patch granularity per region from pixels, ahead of the encoder. VGTok scores each region by multi-scale morphological top-hat separability from its surround, then thresholds those scores at a per-image percentile, which fixes the token budget. A structure-tensor gate ($\lambda_{\min}$) refines only where two-dimensional object structure supports it, leaving one-dimensional clutter coarse. The resulting token set is a strict partition of the image. In a Co-DETR detector with an EVA-02 ViT-L encoder, VGTok clears every published VisDrone-val AP and AP$_S$ at every budget from 40\% to 100\% of tokens. At 40\% it records 44.22 AP with three fifths of the sequence discarded before the first transformer block; dense, it reaches 48.38 AP, $6.08$ above the strongest published entry. VGTok transfers to AI-TOD-v2 untouched, same scorer and same rank, and sets a new state of the art at 37.27 AP and 19.51 AP$_{vt}$. As a pure drop-in into a frozen checkpoint it reaches 36.29 AP at 78.5\% of tokens, above every published entry, where our 376.3M-parameter detector clears a 3.0B multi-expert model. We show that a token budget fixed before the backbone, from local separability and structure geometry alone, holds accuracy on the tiny-object regimes that dominate aerial detection, at $3.1\times$ less encoder compute and $1.9\times$ less encoder memory. Code and models are available at \href{https://github.com/khayrulbuet13/vgtok}{\texttt{github.com/khayrulbuet13/vgtok}} and \href{https://huggingface.co/khayrulbuet13/vgtok}{\texttt{huggingface.co/khayrulbuet13/vgtok}}.

cs.CV↗

arXiv2025

Physics-Guided Surrogate Modeling for Machine Learning-Driven DLD Design Optimization

Sorting cells based on their mechanical properties is essential for applications in disease diagnostics, cell therapy, and biomedical research. Deterministic Lateral Displacement (DLD) devices provide a label-free method for achieving such sorting, but their performance is highly sensitive to cell size and deformability. Designing effective DLD geometries often demands extensive trial-and-error experimentation, as even small variations in cellular mechanical traits can cause significant changes in migration behavior. To address this challenge, we propose a simulation-driven machine learning (ML) framework that predicts suitable DLD design candidates for a given cell type. Our approach integrates high-fidelity particle-based simulations to model cell deformation and migration through microfluidic pillar arrays with supervised ML models trained to estimate optimal geometries. By mapping mechanical parameters such as bending rigidity and shear modulus to deformation index and migration angle, the framework enables rapid, data-informed design of DLD systems. We also demonstrate a deployable web interface to make this tool accessible for real-world device prototyping.

q-bio.QM↗

arXiv2025

Toward Unified Interphase Engineering: The Solid-Electrolyte Interphase in Batteries and Supercapacitors

The development of next-generation electrochemical energy storage requires devices that combine the high energy density of batteries with the power capability and long cycle life of supercapacitors. However, the interfacial phenomena governing performance in these systems remain poorly unified. The solid-electrolyte interphase (SEI), a nanoscale film formed by electrolyte decomposition, is well studied in batteries but its counterpart in supercapacitors has received limited systematic investigation despite growing experimental evidence. This review argues that SEI formation is a universal electrochemical process that occurs whenever electrode potentials drive electron transfer into electrolyte orbitals beyond their stability limits, independent of whether charge storage is Faradaic or non-Faradaic. Differences between battery SEIs and supercapacitor interphases arise mainly from operating conditions, not fundamental chemistry. Engineered interphases created through electrolyte additives, protective coatings, or surface functionalization suppress leakage currents, improve capacitance retention, and enable stable high-voltage operation. By identifying shared mechanisms and establishing transferable design rules, this unified framework provides a foundation for predictive interphase engineering that supports long-lived, high-performance energy-storage technologies.

physics.chem-ph↗

arXiv2025

Curvature-Dependent Polarity of Interfacial Energy Flow in Functionalized CNT Polymer Nanocomposites: A Reactive Molecular Dynamics Perspective

Carbon nanotube (CNT)-polymer composites are widely engineered using surface coatings and chemical treatments to improve interfacial bonding and load transfer. It has been suggested in the nanocomposite literature that nanotube curvature, in conjunction with surface functionalization such as polydopamine (PDA) coating, could serve as an additional control knob for tuning interfacial bonding and energy dissipation in polymer-CNT systems. While experimental and simulation studies have demonstrated the benefits of PDA functionalization, the fundamental mechanism by which nanotube curvature modulates interfacial energy flow and mechanical polarity remains unresolved. This gap is sharpened by a persistent paradox: identical PDA functionalization strengthens some CNT-polymer systems while weakening others, a curvature-dependent inconsistency that has remained unexplained. Here, we employ reactive molecular dynamics (ReaxFF) simulations to resolve how curvature and PDA functionalization jointly govern interfacial energy evolution in CNT-polyvinyl alcohol (PVA) nanocomposites. Our investigation reveals that curvature and PDA functionalization jointly produce opposite regimes of interfacial energy flow: high-curvature CNTs generate dissipative, frictional interphases, whereas low-curvature CNTs confine energy in rigid, cohesive shells. This polarity inversion originates from a curvature-induced transition in PDA adsorption geometry that transforms the interphase from an energy-releasing to an energy-storing configuration. These results establish curvature as a fundamental design parameter for engineering polymer-nanotube interfaces, offering a predictive route to tune interfacial energy flow, mechanical resilience, and transport properties beyond the limits of conventional chemical functionalization.

physics.atm-clus↗

arXiv2025

Real-Time Cell Sorting with Scalable In Situ FPGA-Accelerated Deep Learning

Precise cell classification is essential in biomedical diagnostics and therapeutic monitoring, particularly for identifying diverse cell types involved in various diseases. Traditional cell classification methods such as flow cytometry depend on molecular labeling which is often costly, time-intensive, and can alter cell integrity. To overcome these limitations, we present a label-free machine learning framework for cell classification, designed for real-time sorting applications using bright-field microscopy images. This approach leverages a teacher-student model architecture enhanced by knowledge distillation, achieving high efficiency and scalability across different cell types. Demonstrated through a use case of classifying lymphocyte subsets, our framework accurately classifies T4, T8, and B cell types with a dataset of 80,000 preprocessed images, accessible via an open-source Python package for easy adaptation. Our teacher model attained 98\% accuracy in differentiating T4 cells from B cells and 93\% accuracy in zero-shot classification between T8 and B cells. Remarkably, our student model operates with only 0.02\% of the teacher model's parameters, enabling field-programmable gate array (FPGA) deployment. Our FPGA-accelerated student model achieves an ultra-low inference latency of just 14.5~$\mu$s and a complete cell detection-to-sorting trigger time of 24.7~$\mu$s, delivering 12x and 40x improvements over the previous state-of-the-art real-time cell analysis algorithm in inference and total latency, respectively, while preserving accuracy comparable to the teacher model. This framework provides a scalable, cost-effective solution for lymphocyte classification, as well as a new SOTA real-time cell sorting implementation for rapid identification of subsets using in situ deep learning on off-the-shelf computing hardware.

cs.LG↗

arXiv2023

MIML: Multiplex Image Machine Learning for High Precision Cell Classification via Mechanical Traits within Microfluidic Systems

Label-free cell classification is advantageous for supplying pristine cells for further use or examination, yet existing techniques frequently fall short in terms of specificity and speed. In this study, we address these limitations through the development of a novel machine learning framework, Multiplex Image Machine Learning (MIML). This architecture uniquely combines label-free cell images with biomechanical property data, harnessing the vast, often underutilized morphological information intrinsic to each cell. By integrating both types of data, our model offers a more holistic understanding of the cellular properties, utilizing morphological information typically discarded in traditional machine learning models. This approach has led to a remarkable 98.3\% accuracy in cell classification, a substantial improvement over models that only consider a single data type. MIML has been proven effective in classifying white blood cells and tumor cells, with potential for broader application due to its inherent flexibility and transfer learning capability. It's particularly effective for cells with similar morphology but distinct biomechanical properties. This innovative approach has significant implications across various fields, from advancing disease diagnostics to understanding cellular behavior.

eess.IV↗

arXiv2023

Acoustofluidic Engineering Functional Vessel-on-a-Chip

Construction of in vitro vascular models is of great significance to various biomedical research, such as pharmacokinetics and hemodynamics, thus is an important direction in tissue engineering. In this work, a standing surface acoustic wave field was constructed to spatially arrange suspended endothelial cells into a designated patterning. The cell patterning was maintained after the acoustic field was withdrawn by the solidified hydrogel. Then, interstitial flow was provided to activate vessel tube formation. Thus, a functional vessel-on-a-chip was engineered with specific vessel geometry. Vascular function, including perfusability and vascular barrier function, was characterized by beads loading and dextran diffusion, respectively. A computational atomistic simulation model was proposed to illustrate how solutes cross vascular lipid bilayer. The reported acoustofluidic methodology is capable of facile and reproducible fabrication of functional vessel network with specific geometry. It is promising to facilitate the development of both fundamental research and regenerative therapy.

physics.med-ph↗

arXiv2023

Coarse-Grained Molecular Simulation of Extracellular Vesicles Squeezing for Drug Loading

In recent years, extracellular vesicles such have become promising carriers as the next-generation drug delivery platforms. Effective loading of exogenous cargos without compromising the extracellular vesicle membrane is a major challenge. Rapid squeezing through nanofluidic channels is a widely used approach to load exogenous cargoes into the EV through the nanopores generated temporarily on the membrane. However, the exact mechanism and dynamics of nanopores opening, as well as cargo loading through nanopores during the squeezing process remains unknown and is impossible to be visualized or quantified experimentally due to the small size of the EV and the fast transient process. This paper developed a systemic algorithm to simulate nanopore formation and predict drug loading during extracellular vesicle (EV) squeezing by leveraging the power of coarse-grain (CG) molecular dynamics simulations with fluid dynamics. The EV CG beads are coupled with implicit Fluctuating Lattice Boltzmann solvent. Effects of EV property and various squeezing test parameters, such as EV size, flow velocity, channel width, and length, on pore formation and drug loading efficiency are analyzed. Based on the simulation results, a phase diagram is provided as a design guidance for nanochannel geometry and squeezing velocity to generate pores on membrane without damaging the EV. This method can be utilized to optimize the nanofluidic device configuration and flow setup to obtain desired drug loading into EVs

q-bio.BM↗