Radiomics Analysis of Lymph Nodes with Esophageal Squamous Cell Carcinoma Based on Deep Learning

Figure 2

Selection of radiomic features and deep radiomic features associated with lymph node metastasis via LASSO method. (a) The coefficient profiles of 207 handcrafted radiomic features against the deviance explained. (b) The 10-foldcross-validation curve of handcrafted radiomic features with the optimal lamda value of 0.01151 and 10 nonzero coefficients. (c) The coefficient profiles of 1000 deep radiomic features against the deviance explained. (d) The 10-foldcross-validation curve of deep radiomic features with the optimal lamda value of 0.02024 and 12 nonzero coefficients. (e) The coefficient values of key handcrafted radiomic features. (f) The coefficient values of key deep radiomic features.

(a)

(b)

(c)

(d)

(e)

(f)