Computational ultrasound tissue characterisation for brain tumour resection.
Doctoral thesis (Ph.D), UCL (University College London).
In brain tumour resection, it is vital to know where critical neurovascular structuresand tumours are located to minimise surgical injuries and cancer recurrence. Theaim of this thesis was to improve intraoperative guidance during brain tumourresection by integrating both ultrasound standard imaging and elastography in thesurgical workflow. Brain tumour resection requires surgeons to identify the tumourboundaries to preserve healthy brain tissue and prevent cancer recurrence. Thisthesis proposes to use ultrasound elastography in combination with conventionalultrasound B-mode imaging to better characterise tumour tissue during surgery.Ultrasound elastography comprises a set of techniques that measure tissue stiffness,which is a known biomarker of brain tumours. The objectives of the researchreported in this thesis are to implement novel learning-based methods for ultrasoundelastography and to integrate them in an image-guided intervention framework.Accurate and real-time intraoperative estimation of tissue elasticity can guide towardsbetter delineation of brain tumours and improve the outcome of neurosurgery. We firstinvestigated current challenges in quasi-static elastography, which evaluates tissuedeformation (strain) by estimating the displacement between successive ultrasoundframes, acquired before and after applying manual compression. Recent approachesin ultrasound elastography have demonstrated that convolutional neural networkscan capture ultrasound high-frequency content and produce accurate strain estimates.We proposed a new unsupervised deep learning method for strain prediction, wherethe training of the network is driven by a regularised cost function, composed of asimilarity metric and a regularisation term that preserves displacement continuityby directly optimising the strain smoothness. We further improved the accuracy of our method by proposing a recurrent network architecture with convolutional long-short-term memory decoder blocks to improve displacement estimation and spatio-temporal continuity between time series ultrasound frames. We then demonstrateinitial results towards extending our ultrasound displacement estimation method toshear wave elastography, which provides a quantitative estimation of tissue stiffness.Furthermore, this thesis describes the development of an open-source image-guidedintervention platform, specifically designed to combine intra-operative ultrasoundimaging with a neuronavigation system and perform real-time ultrasound tissuecharacterisation. The integration was conducted using commercial hardware andvalidated on an anatomical phantom. Finally, preliminary results on the feasibilityand safety of the use of a novel intraoperative ultrasound probe designed for pituitarysurgery are presented. Prior to the clinical assessment of our image-guided platform,the ability of the ultrasound probe to be used alongside standard surgical equipmentwas demonstrated in 5 pituitary cases.
Archive Staff Only