Improving Model's Interpretability and Reliability using Biomarkers

  • 2025-01-30 16:55:54
  • Gautam Rajendrakumar Gare, Tom Fox, Beam Chansangavej, Amita Krishnan, Ricardo Luis Rodriguez, Bennett P deBoisblanc, Deva Kannan Ramanan, John Michael Galeotti
  • 0

Abstract

Accurate and interpretable diagnostic models are crucial in thesafety-critical field of medicine. We investigate the interpretability of ourproposed biomarker-based lung ultrasound diagnostic pipeline to enhanceclinicians' diagnostic capabilities. The objective of this study is to assesswhether explanations from a decision tree classifier, utilizing biomarkers, canimprove users' ability to identify inaccurate model predictions compared toconventional saliency maps. Our findings demonstrate that decision treeexplanations, based on clinically established biomarkers, can assist cliniciansin detecting false positives, thus improving the reliability of diagnosticmodels in medicine.

 

Quick Read (beta)

loading the full paper ...