Breast cancer is one of the most diagnosed cancers worldwide, and early detection is essential for improving patient outcomes. When a suspicious lesion is identified through medical imaging, a biopsy procedure is required to determine whether the lesion is benign or malignant. Accurate localisation of the lesion is critical because inaccuracies may result in the lesion being only partially sampled or missed altogether, potentially requiring additional needle insertions, which increases patient discomfort, procedure time, healthcare costs, among others. Ultrasound is widely used because it provides real-time imaging, is relatively inexpensive, and does not expose the patient to ionising radiation. However, conventional image-guided biopsies remain highly dependent on the clinician’s experience, so there is variability in targeting accuracy and procedure outcomes. Robot-assisted biopsy systems aim to address these limitations and provide more consistent probe positioning, controlled needle insertion, and improved integration of information from multiple imaging modalities.
Ultrasound-guided breast biopsy remains challenging because breast tissue deforms during probe manipulation and intervention, causing lesion position and visibility to change over time. These factors can reduce targeting accuracy and increase procedural variability. Recent advances in image registration, tissue-state estimation, and lesion tracking have enabled continuous estimation of lesion location within deformable soft tissue. However, navigation is still largely performed using predefined scanning trajectories, while localisation and tracking information are treated as separate components of the workflow. Consequently, probe motion is not explicitly adapted according to lesion visibility or localisation reliability.
This thesis investigates how lesion localisation and tracking information can be integrated into a robotic control strategy that actively guides ultrasound probe motion during breast interventions. The work is structured around three phases: obtaining an initial lesion estimate suitable for robotic guidance, maintaining lesion localisation and estimating its reliability during tissue deformation, and developing a target-aware navigation framework that uses this information to determine probe motion. Building upon the existing localisation and tissue-state estimation framework, the proposed approach aims to establish a closed-loop connection between lesion estimation and robotic control. By adapting probe motion according to the estimated target state and localisation reliability, the resulting system seeks to maintain lesion visibility and observability throughout the procedure.
