SONATINE

ANR JCJC SONATINE

SONATINE: high-reSolutiON cerebral blood flow estimATIoN basEd on ultrafast ultrasound imaging 

SONATINE is a project founded by French National Research Agency (ANR JCJC) over the period 2024 – 2028.

Project summary

Brain surgery is the usual treatment for most brain cancers. A neurosurgeon partially opens the patient’s skull during surgery, known as a craniotomy, to access the brain. The high-precision removal of the tumor then necessitates an accurate definition of the boundary between the tumor and other vital brain tissues. This assessment is usually done visually by the surgeon with the help of medical imaging techniques, such as ultrafast ultrasound imaging (UUI), a real-time imaging method that has recently become more widespread in clinical practice. Figure 1 shows an illustration of using a SuperSonic Imagine Aixplorer scanner for UUI during a brain tumor surgery that has been recently done at the Neurosurgery department of Tours hospital. Despite the support of this modern imaging tool, the precise demarcation of the tumor boundary remains a complex task, especially for infiltrating gliomas, mainly because of the high vascularization of the peritumoral area, characterized by the proliferation of blood vessels, including tiny vessels with extremely low blood velocities. Accurately determining the microvasculature of the peritumoral area through UUI-based high-sensitivity and high-resolution blood flow estimation is thus gaining interest.

The key idea of blood flow estimation methods is to remove unwanted clutter signals from stationary or moving tissues to reveal blood flow as clearly as possible. Nowadays, the most widely used method formulates a blood-tissue separation model as an inverse problem. More precisely, a UUI data C acquired at a significant high frame (up to several thousands of Hz) is modeled as C = B + T + N, where B is the blood flow, T the tissues, and N the noise. It is called the blood-tissue separation model. Then, a priori knowledge about tissue and blood structures can be incorporated into the blood flow reconstruction process. Usually, the tissues are assumed to be low-rank which is modeled by the nuclear norm ||.||∗, while the blood flow is sparse and promoted by the l1-norm. Moreover, a considerable resolution improvement in the blood flow can be achieved by incorporating a deconvolution using the system point spread function (PSF) H as B = H ⊛ X. This leads to the following deconvolved robust principal component analysis (RPCA) problem:

where ⊛ is the 2D convolution operator, ||.||F is the Frobenius norm, λ and ρ are two hyperparameters balancing the trade-off between the blood sparsity and the tissues’ low-rankness, and X the high-resolution blood flow to be estimated. It is worth noting that the PSF H can be measured experimentally or estimated jointly with the blood flow. Finally, Problem (1) is typically solved using iterative approaches such as the alternating direction method of multipliers (ADMM) or Fast Iterative Shrinkage-Thresholding Algorithm (FISTA). Our preliminary works on RPCA-based methods yielded impressive results both on simulated and clinical data acquired by iBrain team (Tours).

SONATINE is a fundamental trans-disciplinary and transversal research program whose core objectives are:

  • Objective 1: to push the frontiers of the above blood-tissue separation model by taking into account the presence of tissue motion, by deriving better a continuous non-convex relaxation than the current RPCA-based optimization problem, and by making use of advanced Bayesian statistical developments for hyperparameter estimation.
  • Objective 2: to use the resulting techniques to derive a blood flow estimation based on a model- or physics-driven machine learning (ML) approach. Unlike end-to-end learning methods, our approach will take advantage of the flexibility of learning approaches while keeping clear interpretability of their associated model-based imaging schemes. In addition, it will easily handle hyperparameters’ tuning and lead to considerably accurate blood flow estimation.
Figure 1. Operating room of the Neurosurgery department of the Regional University Hospital Bretonneaux of Tours where the external contributors at iBrain have been collaborating for many years. During the surgery, a SuperSonic Imagine Aixplorer scanner was used to perform the UUI modality which helps the surgeon delimit the area infiltrated by the tumor from the healthy brain zones [3].