Preoperative prediction of perineural invasion of rectal cancer based on a magnetic resonance imaging radiomics model: A dual-center study

Figure 3 The process of feature selection and dimensionality reduction through least absolute shrinkage and selection operator. A-C: The process of 10-fold crossvalidation penalty parameter λ selection, according to the rule of the simplest model. The abscissa corresponding to the lowest point of model deviation is the optimal λ; D-F: Each colored line in the variation curve of the characteristic coefficient with value of λ, which was used to determine the non-zero coefficient characteristic parameters based on λ obtained in the upper panel of the figure. T2WI: T2-weighted imaging; T1CE: contrast-enhanced T1WI.