Improving the performance of Depression CAD by using feature selection methods and optimizing system configuration

DOI IR IR (HANDLE) Web Site

Bibliographic Information

Other Title
  • 特徴量選択手法およびシステム構成の最適化によるうつ病CADの高性能化
  • トクチョウリョウ センタク シュホウ オヨビ システム コウセイ ノ サイテキカ ニ ヨル ウツビョウ CAD ノ コウセイノウカ

Search this article

Description

The number of subjects with depression in Japan has been increasing and the number of medical examinees has also been increasing. Therefore, more appropriate diagnosis is required. However, the diagnosis of depression and other mental disorders has not been made using objective indicators such as biomarkers, and the accuracy of the diagnosis has been questioned. Therefore, we are developing a system for calculating the confidence of depression to help physicians to make a diagnosis. To achieve a higher accuracy than the conventional discrimination accuracy, we optimized the program's configuration by adding a voice classifier and using Leave-one-out cross-validation etc. Also, we performed feature selection using 18 feature selection methods and extraction of the optimal model using a brute force algorithm of feature combinations. As a result, the discrimination accuracies of 90 %, 95 %, and 100 % for male, female, and mixed-sex data were achieved, respectively, which is higher than the conventional accuracy of 83 %.

Journal

Keywords

Details 詳細情報について

Report a problem

Back to top