Identification of a Dynamic Model of Pain and Fear Characteristics During Vaginal Dilation Exercises

Treating dyspareunia, i.e., pain during vaginal penetrative sexual intercourse, may include vaginal dilation exercises that are often perceived as uncomfortable (or worse) by patients. Being able to accurately predict the pain and fear levels of these subjects during the treatments is thus in-strumental in designing effective personalized dilation patterns for therapies. Toward this goal, in this paper, we combine an existing qualitative model of vaginal pressure, pain, and fear relations with experimental data obtained during medical trials to derive a parametric model. More precisely, we: 1) analyze how to deal with the identifiability issues caused by the presence of uninterpretable parameters in the original model, 2) use this analysis to derive a novel model that is better suited for data-driven learning purposes, 3) perform a parameter identification using weighted least squares on online and offline measurement data, and 4) test the capability of the overall approach in predicting signals that are proxies of fear and pain levels, comparing the performance one obtains with this refined approach against purely black box Autoregressive moving average exogenous (ARMAX) models. The results indicate that the proposed method works best as a predictive model of fear and pain levels in response to visual and pressure stimuli but still lacks a high level of generalizability.