Computer Assisted Medical Interventions

Computational Photoacoustics

Figure 1: Visualization of the photoacoustic effect: a. Tissue is irradiated by laser light and the laser light diffusely propagates through the tissue; b. where absorbed, the deposited energy leads to the expansion of the tissue which in turn c. leads to the emergence of sound waves that can be measured with an ultrasound transducer.

Photoacoustic tomography (PAT) is a modality that can provide tomographic images at a depth range of up to several centimeters by exploiting the photoacoustic effect (Fig. 1). In the last two decades, a lot of progress has been made towards the translation of PAT into clinical routine and photoacoustics has proven to be an invaluable tool for basic research due to its ability to image functional information, such as local blood oxygenation. Our research contributions can be subdivided into methodological research as well as medical applications.

Challenge: PAI comprises ill-posed inverse problems

Acoustic inverse problem: Measurements of the photoacoustic times series data need to be reconstructed into a spatial distribution of the initial pressure distribution. The main issues that arise are that usually, the reconstruction algorithm is tasked to cope with incomplete data and with a limited detection bandwidth of the measurement device.

Optical inverse problem: A measurement of the initial pressure distribution, even though proportional to the absorption of chromophores, is also dependent on the light fluence which is in turn highly dependent on the distribution of chromophores and scatterers within the tissue. As a consequence, quantification of chromophore concentrations from the underlying initial pressure distribution is an ill-posed inverse problem.

Goal: Quantitative Image Reconstruction

We presented the first machine learning based method towards the optical inverse problem [Kir18c]. Here, a machine learning algorithm uses calculated "context images" to estimate the fluence in each voxel, which – in turn – can be used to correct the optical absorption measured. The patented [A] concept has yielded extremely promising results in silico and thus has high potential for solving one of the primary issues related to PAT.

Complementary to contributions to classical methods to solve the acoustic inverse problem [Kir18a], we have also introduced deep learning concepts for PAT image reconstruction [Wai18] and have made contributions that take into account uncertainty estimation in both inverse problems [Grö18a, Grö18b]. We also developed techniques for 3D photoacoustic vessel angiography with optical tracking [Kir16].

Figure 2: General principle how to extract chromophore concentrations and subsequently calculate functional tissue properties from multispectral photoacoustic images.

Application: Measuring hemodynamic changes in the brain

Our translational contributions in the context of PAT are related to calculation of haemodynamics in the human carotid artery [Kir18b] and open brain surgery [Kir19b], also using deep learning techniques [Grö19].

We used the ability of PAT to recover tissue oxygenation to examine hemodynamic mechanisms while suffering from a stroke. Using PAT to measure such local haemodynamic changes, we were able to visualize waves of spreading depolarization up to a centimeter deep in the brain cortex [Kir19a].

Figure 3: Multispectral photoacoustic (PA) imaging of blood oxygenation (sO2). After a 15min baseline scan, spreading depolarization (SD) was induced by potassium chloride (KCl) stimulation in the left brain hemisphere. The sagittal plane was continuously measured for 51min. (a) PA sO2 estimation before stimulation with marked regions of interest (ROI). (b) Time evolution of estimated sO2 in the two ROI. (c) The change in blood oxygenation (ΔsO2) is shown for three example time steps 30 seconds apart (T1, T2 and T3), corresponding to the dashed lines in (b). Figure and caption reprinted from the CC-BY licensed publication [Kir19a].


  • Dominik Waibel (Masters Student)
  • Angelika Laha (Masters Student)
  • Franz Sattler (Research Assistant)
  • Dr. Thomas Kirchner (PhD Student)

Key Collaborators

  • M.D. Dr. Med. Edgar Santos, Universität Heidelberg, University Clinic for Neurosurgery
  • PD Dr. Med. Sebastian Adeberg, Heidelberg University Hospital at the Department of Radiation Oncology
  • Thomas Held, Heidelberg University Hospital at the Department of Radiation Oncology


[Grö18a] Gröhl, J., Kirchner, T., Adler, T., Maier-Hein, L. Confidence estimation for machine learning-based quantitative photoacoustics. Journal of Imaging (2018).

[Grö19] Gröhl, J., Kirchner, T., Adler, T., Maier-Hein, L. Estimation of blood oxygenation with learned spectral decoloring for quantitative photoacoustic imaging (LSD-qPAI). arXiv preprint arXiv:1902.05839 (2019).

[Grö18b] Gröhl, J., Kirchner, T., Maier-Hein, L. Confidence estimation for quantitative photoacoustic imaging. Photons Plus Ultrasound: Imaging and Sensing (2018), International Society for Optics and Photonics (2018).

[Kir19a] Kirchner, Thomas, et al. "Photoacoustics can image spreading depolarization deep in gyrencephalic brain." Scientific Reports (2019).

[Kir18a] Kirchner, Thomas, et al. "Signed real-time delay multiply and sum beamforming for multispectral photoacoustic imaging." Journal of Imaging 4.10 (2018): 121.

[Kir19b] Kirchner, Thomas, et al. "Photoacoustic monitoring of blood oxygenation during neurosurgical interventions." Photons Plus Ultrasound: Imaging and Sensing 2019. Vol. 10878. International Society for Optics and Photonics, 2019.

[Kir18b] Kirchner, Thomas, et al. "Real-time in vivo blood oxygenation measurements with an open-source software platform for translational photoacoustic research (Conference Presentation)." Photons Plus Ultrasound: Imaging and Sensing 2018. Vol. 10494. International Society for Optics and Photonics, 2018.

[Kir16] Kirchner, Thomas, et al. "Freehand photoacoustic tomography for 3D angiography using local gradient information." Photons Plus Ultrasound: Imaging and Sensing 2016. Vol. 9708. International Society for Optics and Photonics, 2016.

[Kir18c] Kirchner, Thomas, Janek Gröhl, and Lena Maier-Hein. "Context encoding enables machine learning-based quantitative photoacoustics." Journal of biomedical optics 23.5 (2018): 056008.

[Wai18] Waibel, Dominik, et al. "Reconstruction of initial pressure from limited view photoacoustic images using deep learning." Photons Plus Ultrasound: Imaging and Sensing 2018. Vol. 10494. International Society for Optics and Photonics, 2018.


[A] WO2018001702A1 - Machine learning-based quantitative photoacoustic tomography, Lena Maier-Hein, Thomas Kirchner, and Janek Gröhl


  • Best Poster Award 2018 Annual PhD Retreat of the German Cancer Research Center (Janek Gröhl)
  • 1st prize conhIT Nachwuchspreis 2017 for masters thesis "Machine learning based quantitative photoacoustic tomography" (Janek Gröhl)
  • Thomas-Gessmann-Special-Award 2017 for the master thesis "Machine learning based quantitative photoacoustic tomography" (Janek Gröhl)


Keynotes and Invited talks

2020: OSA Biomedical Congress, Machine Learning-based Inference of Functional Tissue Properties from Multispectral Photoacoustic Imaging, Florida, USA (Janek Gröhl)

2019: SPIE BiOS 2019, Deep learning in optical imaging, San Francisco, USA (Lena Maier-Hein)

to top
powered by webEdition CMS