| Summary: | Every year, the healthcare industry collects a huge volume of data that is not mined properly and not put to optimal use. Discovery of the hidden patterns and relationships in data often goes unexploited. Data mining in the medical domain is more rigorous and complex to handle as most available raw medical data are voluminous and heterogeneous in nature. This research mines medical data related to human spine by learning patterns through the collected data and develops medical decision support systems based on intelligent system techniques. This study will help medical specialists in clinical decision making and disease diagnosis related to the spine. The human spine is a multifunctional complicated structure of bones, joints, ligaments and muscles which all undergo change as we age. For most people, these changes occur in a gradual and painless manner. However, a sudden change caused naturally or through injury, can lead to serious medical conditions, which usually result in back pain. Due to the wide diversity of spine functions, any disorder in the spine triggers various severe problems, which negatively affect quality of life and place huge financial and health burdens on the society. While ageing is inevitable, the rate at which the spine shows the effects of ageing is of clinical significance. This research reveals the growth and degenerative pattern of the human spine using intelligent system techniques. The information extracted from lumbar spine MRIs is used to classify age related changes. In this research, principal component analysis was used to detect anomalies in data and to transform the complex multivariate feature space to a smaller meaningful representation. PCA transformation reduced the complexity and dimension of the data, hence permitting a 2D visualization and knowledge of spine growth and degeneration with age. Factor analysis (FA) was used to understand the significance and correlation of spinal features with the normal ageing. Spines were ranked on the basis of their features and clusters were made to group similar samples. Studying the characteristics of the clusters helped in developing an understanding of the variations in spinal features among different age groups. An artificial neural network (ANN) was used in the estimation of age from the extracted lumbar spine features. ANNs have several benefits, including their ability to process complex data, reduced likelihood of overlooking relevant information, and a reduction in the cost and diagnosis time. The ANN model worked well for the spinal age estimation but due to its black box nature, it failed to provide valuable information about the correlations among the spinal features. A hybrid intelligent model consisting of a fuzzy inference system (FIS) and ANN, called an adaptive neuro-fuzzy inference system (ANFIS) was used to extract meaningful information from the data set in terms of fuzzy rules. Self-organizing maps (SOM) were used to visualize variations in lumbar spine features with the natural ageing. Useful information was acquired through SOM exploratory data analysis. Ward and modified Ward clustering methods were employed on SOM to group similar samples and study the characteristics of the clusters. The results from this research are helpful in setting the standards for spinal growth and degeneration with age and for understanding of the spinal disease prevalence. This research will help spine specialists in diagnosing disease from scans. It can be considered as a stepping-stone towards developing a tool for the classification of normal and problematic spines.
|
|---|
