A Model-and-Data Driven Prediction Algorithm on Lumbar Spine Degeneration

The lumbar spine undergoes degeneration over time and is considered the common cause of chronic disease in the lumbar spine. As degradation continues, the chronic disease may develop into severe spinal disease if no proper treatment is provided in time. In this case, a treatment schedule is needed. This study aims to develop a warning system for the degenerative lumbar spine that can assist patients and doctors with treatment schedules. As the data may not hold for Markovian properties, the integration of Model and Data-driven methods is proposed for the system. When Markovian properties hold, we establish an analytical model based on Continuous Time Markov Chain to characterize the degeneration outcome with state distribution probabilities. Otherwise, a data-driven prediction method is utilized to quantify the circumstances of the lumbar spine in patients during their next examination. More specifically, in our machine learning algorithm, indirect prediction methods are proposed to overcome the fact that the absolute value of the indicator may vary too much between consecutive measurements. Meanwhile, to further improve the accuracy, we customize some effective features. Numerical experiments are conducted to demonstrate our approach and validate the contributions.