Compressed All-Optical Photoacoustic Imaging

The question

Photoacoustic imaging has a wonderfully inconvenient premise, send light into tissue, let optical absorption generate ultrasound, detect the sound, and reconstruct where the absorption happened. In principle, elegant. In practice, tissue scatters light, detectors have finite bandwidth, lasers have finite repetition rates, and biology refuses to freeze politely while we measure it.

My PhD asks a simple but annoying question:

Can we recover dynamic biological activity with fewer measurements, using an all-optical photoacoustic system, without throwing away the physics that makes the reconstruction meaningful?

The long-term motivation is neuronal activity imaging. The dream is to observe activity deep inside tissue using optical contrast and acoustic propagation, while avoiding the usual trap of building a beautiful imaging system that is too slow, too dense, or too idealized to survive contact with the experiment.

The approach

I work on compressed all-optical photoacoustic imaging. The system combines compressed optical ultrasound detection, and model-based reconstruction.

The main ingredients are:

• Planar optical ultrasound detection
  The detector is a transparent optical ultrasound sensor, such as a Fabry–Pérot cavity. This geometry is attractive because it can provide broadband ultrasound detection while leaving optical access to the sample.

• Compressed detection with a DMD
  Instead of scanning every point one by one, the detection can be patterned. This opens the door to compressed acquisition, where each laser pulse carries more global information than a single local measurement.

• Inverse problems and sparsity
  The measurements are incomplete by design. The reconstruction therefore has to use prior information: sparsity, positivity, temporal structure, and calcium-like dynamics.

• Fast forward models
  A reconstruction algorithm is only as useful as the model it believes in. I develop shift-invariant planar forward models for photoacoustic propagation, designed to be much faster than brute-force wave simulation while remaining physically meaningful.

• Dynamic reconstruction
  The biological target is not a statue. I study how to reconstruct time-varying activity when the acquisition process is constrained by laser repetition rate, measurement budget, and the general tragedy of finite time.

What came out of it

This is my main PhD project and it is ongoing.

So far, the work has focused on building and validating the computational core:

• fast convolution-based photoacoustic forward and adjoint operators,
• iterative reconstruction methods for planar detection,
• compressed measurement strategies,
• DMD-based compressed detection,
• dynamic reconstruction models for calcium-like signals,
• and experimental alignment/testing toward an all-optical acquisition path.

The main lesson is that reconstruction is not just an algorithmic decoration added after the experiment. It is part of the instrument. The physics, the sampling pattern, the detector geometry, and the reconstruction model all negotiate with each other. Usually rudely.

Why it mattered

This project is the center of my current research identity. It sits exactly at the intersection I care about: optics, acoustics, computational imaging, and inverse problems.

The broader goal is not just to make prettier photoacoustic images. It is to ask what can be measured when the acquisition is deliberately compressed, when the object changes in time, and when the reconstruction model is forced to admit that reality has constraints.

Status

Ongoing PhD project at Institut Fresnel, supervised by Thomas Chaigne and Marc Allain.

Related preprint

Part of the model-based reconstruction work is described in:

A depth-dependent, transverse shift-invariant operator for fast iterative 3D photoacoustic tomography in planar geometry
Ege Küçükkömürcü, Simon Labouesse, Marc Allain, and Thomas Chaigne. arXiv, 2026.