Modeling Touch and Palpation Using Autoregressive Models

Objective: The human haptic system uses a set of reproducible and subconscious hand maneuvers to identify objects. Similar subconscious maneuvers are used during medical palpation for screening and diagnosis. The goal of this work was to develop a mathematical model that can be used to describe medical palpation techniques. Methods: Palpation data were measured using a two-dimensional array of force sensors. A novel algorithm for estimating the hand position from force data was developed. The hand position data were then modeled using multivariate autoregressive models. Analysis of these models provided palpation direction and frequency as well as palpation type. The models were tested and validated using three different data sets: simulated data, a simplified experiment in which participant followed a known pattern, and breast simulator palpation data. Results: Simulated data showed that the minimal error in estimating palpation direction and frequency is achieved when the sampling frequency is five to ten times the palpation frequency. The classification accuracy was 99% for the simplified experiment and 73% for the breast simulator data. Conclusion: Proper palpation is one of the vital components of many hands-on clinical examinations. In this study, an algorithm for characterizing medical palpation was developed. The algorithm measured palpation frequency and direction for the first time and provided classification of palpation type. Significance: These newly developed models can be used for quantifying and assessing clinical technique, and consequently, lead to improved performance in palpation-based exams. Furthermore, they provide a general tool for the study of human haptics.