Machine-learning identified optimised classification models for the diagnosis of typical and atypical lung carcinoids based on the genomic variance

#3972

Introduction: The current diagnosis of lung typical carcinoids (TCs) and atypical carcinoids (ACs) can be ambiguous for small biopsies containing limited tissues.

Aim(s): To aid the diagnosis of lung carcinoids (LCs) by developing classification models based on machine-learning of genomic features.

Materials and methods: The genomic datasets of LCs were retrieved from the American Association of Cancer Research (AACR) Project Genomics, Evidence, Neoplasia, Information, Exchange (GENIE) data repository. In total, 224 LCs including 147 TCs and 77 ACs with 658 genomic features were included. The dataset was divided into a training set and a testing set in a ratio of 7 to 3. Seven machine-learning methods including support vector machine (SVM), decision tree, random forest, gradient boosting decision tree (GBDT), etc. were applied to the training set to generate the classifiers. Least Absolute Selection and Shrinkage Operator (LASSO) was used for feature selection. The performance of each model was evaluated in the testing set.

Conference:

Presenting Author: Guo Y

Authors: Guo Y, Su F, Hu S, Chen R, Chen Q,

Keywords: lung carcinoid, typical carcinoid, atypical carcinoid, machine-learning, diagnosis, genomic features,