Conference Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 551 - Statistical Machine Learning and Artificial Intelligence in Multi-Parametric Quantitative Imaging
Type: Topic Contributed
Date/Time: Thursday, August 11, 2022 : 10:30 AM to 12:20 PM
Sponsor: Section on Statistical Computing
Abstract #323351
Title: Issues in the Development and Validation of Quantitative Imaging Biomarker-Based Models
Author(s): Erich Huang*
Companies: National Cancer Institute
Keywords: high-dimensional data; model validation; simulation studies; medical imaging

The development and validation of a quantitative imaging biomarker (QIB)-based model is riddled with pitfalls. A model developed using error-corrupted QIB measurements will have worse performance than one developed using error-free measurements. QIB model development is highly prone to overfitting, whether by the inclusion of too many QIBs in the model or by fitting an overly complex model form to the data. Improper model validation techniques often produce highly biased estimates of the model performance; even model validation using some proper techniques may lead to problems in small sample size scenarios.

An overview of the types of problems one may encounter during the development and validation of a QIB-based model is presented and demonstrated through simulation studies. Loose recommendations on model development and validation suggested by these results are then given.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2022 program