Comprehensive Summary
Data was obtained from six Yucatan minipigs at time points: baseline (before surgery), weeks 4, 8, 12, and 16 after surgery. Quantitative Chemical exchange saturation transfer (qCEST) imaging was then performed on the porcine models. Magnetization transfer ratio (MTR) and exchange rate features were then derived from the CEST data to be used for training of the PRF model. The Glasgow Pain Scale and (wind-up-ratio) WUR assessment were used to assess general discomfort, mobility, behavioral responses, pain sensitivity, and central sensitization. 3D SS-CEST multitasking was performed alongside the conventional 2D CEST generation. The data showed a 81% correlation between SS-CEST and conventional 2D CEST, with a nearly 22 factor decrease in time spent processing. There was also an accuracy rate of 80% in prediction of pain scores. Overall, this paper presents the potential of non-linear machine learning assessments and the potential of SS-CEST in clinical practice. However further research is needed to generalize towards human populations with more variables than compared to porcine models.
Outcomes and Implications
This research is valuable for the clinical use of qCEST, particularly in assessment of non-invasive pathologies such as lower back pain. Despite the issues with generalizability, this research also highlights the value of machine learning in both academic and clinical settings for assessing diagnostic imaging data and non-linear interactions between features. Given time for machine learning research in more areas of clinical practice, there is a vast potential for improving patient care in chronic pain related pathologies.