DBDeepan Balakrishnan

Deepan Balakrishnan

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Deepan Balakrishnan has a total of 47 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Nanoscale cuticle mass density variations influenced by pigmentation in butterfly wing scales

    by Deepan Balakrishnan, Anupama Prakash, Benedikt J. Daurer, Cédric Finet, Ying Chen Lim, Zhou Shen, Pierre Thibault, Antónia Monteiro, N. Duane Loh

    How pigment distribution influences the cuticle density within a microscopic butterfly wing scale, and how both impact each scale’s final reflected color, remains unknown. We use ptychographic X-ray computed tomography to quantitatively determine, at nanoscale resolutions, the three-dimensional mass density of scales with pigmentation differences. By comparing cuticle densities between two pairs of scales with pigmentation differences, we determine that the density of the lower lamina is inversely correlated with pigmentation. In the upper lamina structure of Junonia orithya and Bicyclus anynana, low pigment levels also correlate with sheet-like chitin structures as opposed to rod-like structures. Within each scale, we determine that the lower lamina in all scales has the highest density, and distinct layers within the lower lamina help explain reflected color. We hypothesize that pigments, in addition to absorbing specific wavelengths, can affect cuticle polymerization, density, and refractive index, thereby impacting reflected wavelengths that produce colors.


  • Open Access

    Using X-ray Spectro-ptychography to Unveil the Structural Origins of Colour on Butterfly Scales

    by Justin Ong, Benedikt J. Daurer, Anupama Prakash, Deepan Balakrishnan, N. Duane Loh

    When an object is illuminated by a source, the resulting exit wave is a complex function intricately connected to the object’s structure. Generally speaking, the wave’s modulus corresponds to the object’s transmittance, while its phase represents the cumulative phase shift relative to free space, caused by the object’s refractive index as the wave traverses it. Therefore, to fully characterise the object, both the modulus and phase of the exit wave are required. However, detectors only measure the intensities (the modulus squared) of the waves striking the detector. The modulus can thus be easily obtained by taking the square root of the measured intensity. The phase, on the other hand, is much more challenging to determine. Any potential phase configuration could theoretically produce the same measured intensity, leading to what is known as the ‘Phase Retrieval Problem’, whereby the true phases that were lost during the experiment are not easily recovered. Ptychography offers a possible solution to the phase retrieval problem [ 1]. Multiple diffraction intensity patterns are obtained by scanning a finite probe over an extended specimen with sufficient overlap between adjacent illumination positions. Combined with suitable iterative reconstruction algorithms, these measured diffraction patterns can be effectively utilised to accurately determine both the amplitude and phase of the object’s exit wave, providing a robust approach that significantly enhances the ability to retrieve complete structural information about a specimen. Ptychography can be combined with X-ray absorption spectroscopy by using an X-ray beam of tuneable energies at each scan position. The diffraction patterns are now collected when the X-rays are scanned over absorption edges of specific elements, providing chemical information, in addition to high resolution reconstructions.

  • Open Access

    Data-Driven Analysis of STEM-EELS Spectra for High-Resolution Compositional Mapping

    by Zhen Yuan Yeo, Wenhui Lai, Jong Hak Lee, Deepan Balakrishnan, Barbaros Özyilmaz, N. Duane Loh

    Scanning Transmission Electron Microscopy coupled with Electron Energy Loss Spectroscopy (STEM-EELS) is a powerful technique that reveals detailed information about the local chemical structure within a sample. With advancements in spatial-resolution, STEM-EELS now enables the detection of fine structural differences, such as distinguishing between sp 2 and sp 3 bonding in amorphous carbon [ 1], making it invaluable for studying complex materials. However, the high dimensionality and noise inherent in STEM-EELS data present significant challenges in data interpretation and compositional mapping. Traditional methods often rely on reference spectra, which can limit the identification of novel or unexpected phases. In this work, we present an analytical workflow that enhances compositional mapping and structural identification by integrating advanced dimensionality reduction, noise filtering, and clustering techniques. This approach not only improves spectral clarity and spatial resolution but also opens new possibilities for discovering unknown material phases. Here, we demonstrate a comprehensive analytical workflow designed to address the challenges inherent in high-dimensional STEM-EELS data. STEM-EELS was applied on a sample of Silicon-Carbon composite in battery anode material ( Fig 1A-B) and a control sample (without chemical bonding). Our approach begins with Principal Component Analysis (PCA)-based noise reduction ( Fig 1C), which significantly improves spectral clarity by isolating subtle elemental features that are often masked by noise. To manage the complexity of the data, we then apply t-distributed Stochastic Neighbor Embedding (t-SNE) for dimensionality reduction, preserving local spectral relationships and enabling the visualization of intricate compositional patterns. K-means clustering is subsequently employed on the t-SNE output ( Fig 1D), allowing us to partition the data into distinct compositional regions and facilitating phase identification across the sample. Figure 1: Schematic of data-driven analysis of STEM-EELS Spectra to detect subtle difference in chemical compositions within Si-C composites. A) Thickness contrast of Silicon-Carbon sample at the Carbon K-edge. B) A sample of EELS spectra at various energy channels. C) PCA noise-reduction on EELS spectra. D) t-SNE embedding of EELS spectra with K-means clustering. E) Significant regions with composites of Si, C, O are obtained from mutual information. F) Overlay of significant composites shows spatial dependence which implies formation of compounds, rather than mixtures. G) Vector quantization of significant clusters of carbon shows a change in material composition from the control sample (decrease in sp 2 bonds).

  • Open Access

    Ghostbuster: Beyond Ewald Sphere Curvature Correction for Cryo-EM Single Particle Reconstruction Via Phase Retrieval

    by Joel Yeo, Benedikt J. Daurer, Dari Kimanius, Tristan Bepler, Deepan Balakrishnan, Yong Zi Tan, N. Duane Loh

    Recent advances in single particle analysis (SPA) for cryogenic electron microscopy (cryo-EM) have significantly enhanced the resolution of 3D particle reconstructions. However, a gap remains between the theoretical models of cryo-EM image formation and experimental results. While the multislice and in-line holography models accurately describe image formation in electron microscopy experiments [ 1], most cryo-EM SPA software still rely on simplified linear approximations, like projection approximation and contrast-transfer function theory. These linearized models lead to ghost-like artifacts ( Fig. 1) in the reconstructed particle due to the inaccurate Fourier coefficients inferred from phase-less cryo-EM images, which are subsequently used in Ewald sphere curvature correction (ESCC) algorithms. As such, current approaches cannot fully remove the effects of multiple scattering in thick particles [ 2]. In this work, we aim to close this theory-experiment gap by incorporating the full physics of the imaging process to go beyond Ewald sphere curvature correction (ESCC) algorithms for cryo-EM single particle reconstruction. To achieve this, we developed a diffraction tomography algorithm, Ghostbuster, which refines 3D particle reconstructions using batch stochastic gradient descent. Ghostbuster minimizes the error between simulated images from a multislice-based forward model and actual cryo-EM measurements. Implicitly, Ghostbuster retrieves the lost phases of the measured images that are then backpropagated to reconstruct the 3D particle. We validate our approach by comparing the reconstruction of an adeno-associated virus serotype 2 variant (AAV2) particle (Empiar-10202) between Ghostbuster and CryoSPARC, a state-of-the-art cryo-EM processing software. We demonstrate a reduction in ghost-like artifacts in Ghostbuster’s reconstruction, which leads to an overall improvement in resolution based on both half-map and map-to-model Fourier shell correlation (FSC) metrics ( Fig. 2). Fig. 1. (Left) A ghost atom appears due to interference from two atoms at z = 0 after some distance away. (Right) No ghost due to projection approximation. Fig. 2. Comparison of 3D reconstruction for AAV2 between CryoSPARC (with and without ESCC) and Ghostbuster. (a) Local resolution estimate 3D renders. (b) The 128 th slice from the model and reconstruction 3D particles. (c) Both half-map and map-to-model FSCs of the reconstructions.

  • Open Access

    Correlative Imaging Provides Novel Insights into Butterfly Wing Scales’ Coloration, Structure, and Density

    by Deepan Balakrishnan, Anupama Prakash, Benedikt J. Daurer, Cédric Finet, Ying Chen Lim, Antónia Monteiro, N. Duane Loh

    The butterfly wing scales are biocomposites with impressive structural and optical properties, owing to their features spanning from hundreds of microns to a few nanometers. Though they demonstrate efficient color production and structural rigidity, the mechanisms to achieve them are well-concealed across different length scales. As a result, understanding these biocomposites' composition, structure, and corresponding function remains challenging. No single technique can provide a high-resolution three-dimensional (3D) volume with an accurate color map. Traditionally, approaches such as light microscopy (LM) and spectrophotometry for color responses, and scanning electron microscopes (SEM) and transmission electron microscopes (TEM) for structural imaging provide key insights [ 1]. However, these methods offer either limited resolution (LM) or limited field of view (TEM), falling short of capturing the full complexity of the scales [ 2]. Here, we used ptychographic X-ray computed tomography (PXCT), which provides 3D density maps at an intermediate resolution (66.5nm) over hundreds of microns. This allowed us to combine the inferences from the high-resolution (1 μm) color response from LM/spectrophotometry [ 3, 4]. In Figure 1, the images obtained from various modalities are shown along with their corresponding length scales. Notably, PXCT provides both 3D structural measurements and density values. The 3D structural measurements enable us to estimate the tilt map of the continuous lower lamina structure of the scale and the thickness map of the intricate upper lamina structure. The known physics behind the reflectance and absorbance spectra enables us to combine insights from a wide range of imaging modalities synergistically. Fig. 1. Butterfly wing scales have features spanning from 100s of microns to a few nanometers captured using different imaging modalities. From left to right: 1, light microscope (LM), 2, ptychographic X-ray computed tomography (PXCT), scanning electron microscope (SEM), focused ion beam SEM (FIB-SEM), transmission electron microscope (TEM), resin embedding TEM (RE-TEM). The scale bar represents the corresponding length shown below each image.

  • Open Access

    Reaching the Limits of Phase Contrast in Transmission Electron Microscopy: Bridging the Theory-Experiment Gap With Physics Informed Neural Networks

    by Joel Yeo, Yixiao Yang, Benedikt J. Daurer, Deepan Balakrishnan, N. Duane Loh

    Advances in high-resolution electron microscopy have enabled remarkable insights into both biological and materials science, particularly in the detailed structural imaging of small particles and complex materials interfaces. Yet, as resolution increases, the effective depth of field (DOF) of Transmission Electron Microscopy (TEM) narrows, creating new challenges for accurately imaging thick samples. Traditional phase retrieval approaches rely on contrast transfer functions (CTF) and projection approximations (PA) that linearize image formation physics. While effective at near-atomic resolutions, these methods lose critical phase information, limiting resolution in cases that require a full 3D reconstruction of complex structures. The maximum achievable contrast in Transmission Electron Microscopy (TEM) for small particles, including biomolecules, is theoretically set by the scattering cross sections of these particles. Our previous work has shown that, under ideal conditions, computational phase retrieval can reach this theoretical limit by precisely modeling electron-matter interactions 1 . However, the real-world limitations of TEM—such as noise, sample damage, and heterogeneity in imaging conditions—create a gap between theoretical predictions and actual experimental outcomes, especially for challenging samples like small or flexible biomolecules. This theory-experiment gap has historically hindered the ability to fully exploit TEM for resolving fine structural details in complex materials. First-principles simulations struggle to account for the empirical effects present in real data, including variations in ice thickness, specific particle damage, and environmental noise. To address this gap, we integrate these theoretical models into Physics-Informed Neural Networks (PINNs), which incorporate the physical laws governing electron scattering directly into the machine learning framework.

  • Open Access

    High-Throughput and Dynamic 3D TEM Imaging with Iterative Pop-out Metrology

    by Deepan Balakrishnan, Joel Yeo, Michel Bosman, Utkur Mirsaidov, N. Duane Loh

    Advancements in nanofabrication, semiconductor manufacturing, battery science, and biological processes have immensely benefited from in-situ transmission electron microscopy (TEM), which reveals dynamics in intermediate states. However, our understanding remains limited to two dimensions, as the tomographic techniques require multiple tilts for 3D reconstruction, extending acquisition time and reducing time resolution, making it challenging to capture rapid 3D structural dynamics in the millisecond range, which is critical for understanding failure modes in nanofabrication and self-assembly mechanisms that occur in nature [ 1]. Our recent work, pop-out 3D metrology, is a computational optics technique that exploits the physics of wave-specimen interaction and known material priors to create a 3D image from a single 2D projection [ 2– 3]. This technique applies broadly to coherent bright-field imaging with electrons, photons, or other wavelike particles. However, pop-out metrology requires high signal-to-noise, assumes the specimen is homogenous, and produces artifacts for multi-layered complex specimens. Hence, we are extending the pop-out 3D imaging for TEM movies for complex specimens. Here, we employ a two-stage method; first, we generate initial reconstructions from the pop-out 3D metrology. In the second stage, a physics-based multi-slice forward model is used to refine the reconstructions iteratively. The reconstructions are updated to minimize the error between the experimentally captured TEM images and the model output with an additional regularizer that constrains and smooths the reconstructions across the frames. Our preliminary result shows reconstruction for a simulated TEM movie of a particle (nanopillar) dry-etching in a gas cell enclosed by 20-nm thick upper and lower membranes ( Figure. 1). Fig. 1. Proof-of-concept: etching 3D dynamics of a simple dual-layered structure (top row) can be inferred from a time series of their single 2D simulated TEM images (bottom row).

  • Open Access

    Using Pop-Out 3D Metrology to Image Large Areas and Fast Dynamics to Nanometer Resolutions

    by Deepan Balakrishnan, Joel Yeo, Zhaogang Dong, Ramon Paniagua-Dominguez, Michel Bosman, Utkur Mirsaidov, Duane Loh

    Emerging technologies such as meta-surfaces, nano-catalysis, and solid-state batteries now lean toward fabricating more complex 3D structures. To understand these 3D nanostructures and to guide their fabrication, the lack of rapid 3D metrology remains a critical blind spot. Though transmission electron microscopes (TEM) are the most prominent tool for investigating materials at the nanoscale resolution, they are predominantly used for 2D measurements. Furthermore, routine inspection and defect analysis of fabricated 3D nanostructures in multi-layered planar microelectronics is still limited to 2D.


  • Open Access

    Ghostbuster: A phase retrieval diffraction tomography algorithm for cryo-EM

    by Joel Yeo, Benedikt J. Daurer, Dari Kimanius, Deepan Balakrishnan, Tristan Bepler, Yong Zi Tan, N. Duane Loh


  • Open Access

    Single-shot, coherent, pop-out 3D metrology

    by Deepan Balakrishnan, See Wee Chee, Zhaslan Baraissov, Michel Bosman, Utkur Mirsaidov, N. Duane Loh

    Three-dimensional (3D) imaging of thin, extended specimens at nanometer resolution is critical for applications in biology, materials science, advanced synthesis, and manufacturing. One route to 3D imaging is tomography, which requires a tilt series of a local region. However, capturing images at higher tilt angles is infeasible for such thin, extended specimens. Here, we explore a suitable alternative to reconstruct the 3D volume using a single, energy-filtered, bright-field coherent image. We show that when our specimen is homogeneous and amorphous, simultaneously inferring local depth and thickness for 3D imaging is possible in the near-field limit. We demonstrated this technique with a transmission electron microscope to fill a glaring gap for rapid, accessible 3D nanometrology. This technique is applicable, in general, to any coherent bright field imaging with electrons, photons, or any other wavelike particles.