Assessment of Clinical Variables Importance with the Use of Neural Networks by the Example of Thyroid Blood Test Parameters
[ 1 ] Instytut Mechaniki Stosowanej, Wydział Budowy Maszyn i Zarządzania, Politechnika Poznańska | [ P ] employee
2019
chapter in monograph / paper
english
- hypothyroidism
- neural network analysis
- parameter importance
EN Screening blood tests for thyrometabolic status determination are difficult in interpretation because of many factors like age, sex or measurement method that influence their proper interpretation. To solve this problem machine learning techniques, like artificial neural networks (ANN) can be applied, but their application is very common belittled due to their very little explanatory insight to the relation between input parameters (e.g. test results) and model of the disease. In contrast to previous studies concerning application of neural networks in thyroid disease diagnosis, in this study the authors decided to focus on extraction of reliable dataset (with preserved proportion of diseased and health cases) and quantification of input parameters importance in neural network decisive process. The importance of the variables considered as the most significant in hypothyroidism detection was estimated based on two independent methods: connection weights method (according to the Garson’s algorithm) and sensitivity analysis. The results show, that the most important factors in hypothyroidism detection are TSH, TT4, FTI and age, and the rejection of other analyzed in this study parameters (sex, T3, T4U) does not influence significantly the performance of the neural network model and its predictive power.
15.08.2019
36 - 46
20