Mapping Tactile Imaging Information: Parameter Estimation and Deformable Registration

author:Anna M Galea
adviser:Robert D. Howe
degree: Ph.D.
institution: Harvard University

Medical Tactile Imaging uses an array of pressure sensors mounted on a rigid scanhead to record the surface pressures that result when the scanhead is pressed into biological tissue. The resulting tactile data quantifies palpation, and contains information on the stiffness of the underlying tissue as well as the geometric distribution of the stiffness. Tactile imaging shows promise for clinical use in breast palpation and in assessing tissue properties in organs such as the liver. To date, tactile information has been used to estimate tissue geometry but not stiffness. We develop a linear algorithm to estimate the salient tissue parameters from a simple model of a solid lesion attached to the substrate of soft tissue. The parameters of interest are the background stiffness and thickness, and the stiffness and diameter of a round lesion. The algorithm is developed using finite element models, and results obtained on physical models show errors of 5.4% in estimating lesion modulus. This work was extended to the case of a solid lesion floating in soft tissue, which encompasses cases of pathology in a large breast or pathology in the liver or prostate. Parameter estimation from finite element data showed errors of 12% for the modulus of large lesions. Extending this model to hollow areas in soft tissue, such as the large veins in livers, resulted in errors of 25 and 13% for estimating the size and depth of veins in perfused porcine livers. Given the success of applying a linear algorithm to the relationship between tissue parameters and tactile information, we study the impulse response of the system to explore the limits of tactile imaging using the currently available scanheads. Employing tactile imaging clinically in breast cancer screening requires registration of tactile images to other modalities such as mammograms. A deformable registration algorithm is developed on finite element models and applied to a physical model with less than 2.4 mm registration error. A preliminary clinical study shows good registration between the tactile and mammographic images, and holds promise for increasing the positive predictive value of breast cancer screening.

Harvard BioRobotics Laboratory Home