Chandrasekar
SUBRAMANI NARAYANA

Research focus

Computer vision for microscopy.

My work follows a continuous question across imaging modalities: how can software convert complex visual information into measurements that scientists and engineers can trust?

01

Microscopy computer vision

Designing algorithms that detect, segment, count, and quantify structures in scientific images.

02

Quantitative phase imaging

Developing software around label-free phase information in my current engineering work at Phasics.

03

Reliable scientific AI

Connecting learning-based predictions to physical models, statistics, and reproducible validation.

How my current instruments see

Quantitative phase imaging, from a wavefront's point of view.

Phasics' sensors are built on quadriwave lateral shearing interferometry (QWLSI), a common-path, self-referencing technique that needs no separate reference beam. Step through the sequence, then draw your own sample below to see it measured.

01 · Sample

Try it yourself

Draw a sample, watch the measurement respond.

Paint a phase object below, brighter regions read as optically "thicker". The 3D sample and reconstruction above update to match, alongside what a plain camera and a QWLSI sensor would each actually see.

Draw a sample (drag to paint)

Interferogram (at the sensor)

Intensity (plain camera)

Nearly invisible. A phase object barely changes brightness.

Phase (QWLSI recovers it)

low OPDhigh OPD

The core relation (simplified)

∇W(x,y) ≈p2πd· ∇φ(x,y)

The fringe pattern's local phase shift φ gives the wavefront's gradient, scaled by grating period p and grating–sensor distance d.

W(x,y) =∇W · d𝓁

Integrating the two gradient maps (x and y) recovers the full quantitative phase map W, the optical path difference at every pixel.

W wavefront / optical path difference · φ measured interferogram phase · p grating period · d grating–sensor distance

Methods

Built for reproducibility and interpretation.

My toolkit includes classical computer vision, segmentation, density-map estimation, detection and classification, PyTorch and TensorFlow, C++ and Python scientific software, synthetic-data design, statistical modelling, and HPC/SLURM workflows.

During my PhD I applied these methods to fluorescence and TIRF microscopy. At Phasics, my imaging focus has expanded to quantitative phase images and the engineering questions surrounding their analysis.

Across modalities, I care about transparent validation, controlled experiments, clear visualisation, and software that another person can understand and maintain.

See methods in practice ↓ Explore imaging evolution →

Selected projects

Research translated into working systems.

Case studies spanning quantitative microscopy, biological image analysis, spectroscopy, medical imaging, and scientific software.

Interactive laboratory

Explore the ideas in motion.

These demonstrations translate microscopy acquisition and computer-vision analysis into interactive visual explanations.

Interactive microscopy desk

Different images.
The same need to measure well.

I have worked from labelled fluorescence and surface-selective TIRF data during my PhD to quantitative phase images in my current work at Phasics. Move across the field and switch modalities to explore the visual logic of each.

50 µm

Intensity reveals labelled structures and events.

Inside the computer-vision pipeline

From pixels to quantitative predictions.

The visualisation separates work completed during my PhD from newer foundation-model approaches I am exploring conceptually for microscopy.

Animated density-map neural networkPoint annotations pass through encoder and decoder layers to produce continuous density peaks.POINT ANNOTATIONSU-NET REPRESENTATIONDENSITY MAP
TRAINING EPOCH001 / 120
Completed PhD work

Learning a density map from point annotations

A U-Net-style model learns a continuous density field from sparse bacteriophage locations. Integrating the predicted map returns an interpretable count while retaining spatial information.