Patient-Specific Brain Networks from BOLD-fMRI

­Summary: A method for analysing functional magnetic resonance (fMRI) images to identify brain networks in task and resting-state experiments.
Blood-oxygen-level dependent functional MRI (BOLD-fMRI) measures change in neuronal activity by detecting alterations in brain blood oxygenation. High oxygenation levels are associated with increased blood perfusion, and ultimately increased neuronal activation. The proposed technology allows to process time-varying BOLD-fMRI scans to obtain patients-specific brain networks and their corresponding neural activation profiles during task and rest. Specifically, the technology allows to simultaneously decompose observed BOLD signals into maps of spatial position, neuronal activity time courses, and hemodynamic responses, that can be used in pre-operative planning of brain surgery.

BOLD-fMRI seeks to map brain areas based on cortical activity. BOLD-fMRI is frequently used in studies of the brain to allow the identification and analysis of functional areas of the eloquent cortex (motor, language, audio and visual processing).

However, BOLD-fMRI techniques are not able to identify unexpected subject behaviour, with the associated risk of leading to an erroneous assessment.

Hence, a robust method to obtain estimates of brain networks through BOLF-fMRI has been a long-felt need.


The proposed technology allows to overcome the above-mentioned limitations, by providing means to obtain robust and spatially specific estimates of brain networks in a single patient.

Specifically, the proposed technique can reveal individual task scans where subjects did not comply with task instruction. This is a strong advantage over other known analysis models applied to fMRI data because unexpected subject behaviour (such as depicted in the Figure red rectangular shown below) leads to the derivation of wrong spatial activation maps. Retrospective analysis of task experiment scans can therefore reveal non-compliant task participation in circumstances in which task fMRI is currently in use for surgery planning.

Ultimately, the proposed technique enables reliable spatial brain mappings, which translate into more cost-effective and less invasive planning of epilepsy neurosurgery and brain tumour resection.

The proposed technique may therefore offer a non-invasive cost-efficient alternative to electric stimulation assessment of brain function for pre-operative planning in brain surgery.

The technology is protected by a US patent application and a European patent application and is available for licensing. Suitable commercial partners are sought for further development and commercialisation. 
The Science
The technology is based on a neurobiologically-driven matrix factorisation approach applied to BOLF-fMRI time varying data, by modelling the acquired data as the convolution of a neural activity time course with a haemodynamic filter. Multiple tasks activating several parts of the eloquent cortex can be conducted in one BOLD-fMRI experiment design lasting for 5-15 minutes.


Figure: Each row depicts task blocks (lines in four distinct colours in row one to four) of either foot, left hand, right hand or tongue movement, respectively. The neural activation time course of the spatial map (left column) with the highest correlation to a particular movement type is depicted as continuous blue line superimposed on the individual block movement timings. The activation time course in the red rectangle depicts subject 6 confusing left with right hand movement.

IP Status
The technology is protected by:
EP3803427A1 EP Pending Application; and
US 17/058,714 US Pending Application.
Further Information
Hütel, M., Melbourne, A., Ourselin, S., 2018. "Neural Activation Estimation in Brain Networks During Task and Rest Using BOLD-fMRI", Lecture Notes in Computer Science, 11072, pp. 215–222. doi:10.1007/978-3-030-00931-1_25
Patent Information:
For Information, Contact:
Lorenza Grechy
King's College London
Sebastien Ourselin
Michael Hutel