.

In this project, together with the company PA Imaging, Insyte Technology, the University of Twente, and RadboudUMC, we are collaborating on 3D PAUS: Elastography for faster Detection of Breast cancer in 3D Photoacoustic Mammography. We are pleased to have commenced this joint effort and appreciate the trust placed in our organisation by EFRO.

Tumor formation is characterized by the formation of new blood vessels (angiogenesis). Breast cancer can be detected by recognizing angiogenesis among the bulk of healthy blood vessels. Painless 3D-PAUS, which stands for “3-dimensional photo-acoustic ultrasound”, is a completely new mammography method that allows the blood vessels in the breast to be visualized. However, it turns out that it is not so easy for the radiologist to recognize (early) angiogenesis purely by visual inspection. To expedite this recognition of angiogenesis, the partners within this project intend to expand an existing 3D-PAUS mammography system (PAM3+) with elastography.

With elastography, the breast is set into controlled, minimal vibration. This is painless. Because tumor tissue is harder and firmer than the surrounding tissue, when the breast vibrates, the blood vessels around the rigid tumor will move slightly differently from the surrounding healthy tissue. This subtle difference in movement can be made visible as a stroboscopic effect using the same laser light flashes applied in photoacoustic mammography. Making the difference in tissue stiffness visible stroboscopically in this way helps the radiologist more easily recognize angiogenesis among the rest of the healthy vascular system.

In addition to elastography, artificial intelligence will be developed to recognize angiogenesis between healthy blood vessels. This innovation and this project directly address the demand to detect breast cancer in a better, more efficient, and, above all, more woman-friendly manner.

Workpackage: Hybrid model- and data-driven neural operators for acoustic inverse problems

Neural networks, the backbone of artificial intelligence, have recently been increasingly used for scientific problems. Particularly, it has been applied to solving inverse problems in medicine and geophysics. Another recent advancement is the neural operator. These neural networks are able to map functions to functions and are mainly used in scientific computing as fast surrogate models for solving complex PDEs.

In inverse problems involving partial differential equations (PDEs), these neural operators can be very important as they are much quicker in solving the PDEs and hence can help in solving the inverse problem faster. One issue, however, is that the neural operators are mainly data-driven. Hence, there are no guarantees that the PDE is solved accurately. This is in contrast to solving the PDE via (slower) model-driven numerical methods. The lack of guarantees is a big issue for neural operators, as inaccurate PDE solutions can have a significant negative impact on the obtained solution.

This project tries to tackle this problem for inverse problems in (photo)acoustics. Particularly, we want to combine the model-driven approach with the data-driven neural operators. The goal is to develop a general method that provides improved and faster reconstructions to the inverse problem by using neural operators, but that has some additional guarantees by incorporating model-driven components, hence creating a hybrid model.

 From the paper:
“A Mathematical Guide to Operator Learning” by Boullé and Townsend.

From the paper:
“High resolution 3D ultrasonic breast imaging by time-domain full waveform inversion” by Lucka et al.