A Machine Learning Based Model to Identify The Relative Location of CT Scan Slices

Background and aims: Registering CT scans in a body is an important technique for aligningand comparing different CT scans. In this method required for navigating automatically to certainregions of a scan. In, a unique correspondence is created between one point of one image andanother point of the second image in such a way that both of them represent the same point of theimage. Considering the large number of slices of CT scan images. This method is one of the newmethods for aligning and comparing different CT scan images. The purpose of this study is to providea model based on machine learning algorithms to identify the relative location of CT slices.Method: In this study, the data set of UCI database titled relative location of CT slices was used.۵۳۵۰۰ CT scan images belonging to ۷۴ patients (۴۳ men and ۳۱ women) were used to build theprediction model. Each CT scan slice was described by two histograms in polar space, whichshowed the position of the skeletal structure and the air inclusions inside of the body. The targetvariable was a number between ۰ (top of the head) and ۱۸۰ (soles of the feet) that indicated therelative position of an image. Considering that the target variable were continuous, two ML methodsincluding linear regression and artificial neural network (multilayer perceptron: MLP) wereused to build the prediction model. Prediction models were evaluated by ۱۰-fold cross validationand are implemented in MatLab environment.Results: The results showed that MLP with Lunberg-Marquardt training algorithm with R۲ andRMSE values equal to ۰.۹۹۴۷ and ۲.۲۶۱ have better performance compared to the linear regressionmethod with R۲ and RMSE values ۰.۹۲۸۶ and ۸.۲۹۱ respectively.Conclusion:The results revealed that the neural network prediction model achieved to betterperformance compared to the linear regression method for predicting the relative location and thesame areas in the CT slices.

کلیدواژه ها:

image registration ، CT scan ، machine learning ، artificial neural network ، linear regression