Faghani Shahriar, Khosravi Bardia, Zhang Kuan, Moassefi Mana, Jagtap Jaidip Manikrao, Nugen Fred, Vahdati Sanaz, Kuanar Shiba P, Rassoulinejad-Mousavi Seyed Moein, Singh Yashbir, Vera Garcia Diana V, Rouzrokh Pouria, Erickson Bradley J
Radiology Informatics Laboratory, Department of Radiology, Mayo Clinic, 200 1st St SW, Rochester, MN 55905.
Radiol Artif Intell. 2022 Aug 24;4(5):e220061. doi: 10.1148/ryai.220061. eCollection 2022 Sep.
The increasing use of machine learning (ML) algorithms in clinical settings raises concerns about bias in ML models. Bias can arise at any step of ML creation, including data handling, model development, and performance evaluation. Potential biases in the ML model can be minimized by implementing these steps correctly. This report focuses on performance evaluation and discusses model fitness, as well as a set of performance evaluation toolboxes: namely, performance metrics, performance interpretation maps, and uncertainty quantification. By discussing the strengths and limitations of each toolbox, our report highlights strategies and considerations to mitigate and detect biases during performance evaluations of radiology artificial intelligence models. Segmentation, Diagnosis, Convolutional Neural Network (CNN) © RSNA, 2022.
机器学习（ML）算法在临床环境中的使用日益增加，引发了人们对ML模型偏差的担忧。偏差可能出现在ML创建的任何步骤中，包括数据处理、模型开发和性能评估。通过正确实施这些步骤，可以将ML模型中的潜在偏差降至最低。本报告重点关注性能评估，并讨论模型适用性以及一组性能评估工具箱：即性能指标、性能解释图和不确定性量化。通过讨论每个工具箱的优缺点，我们的报告强调了在放射学人工智能模型性能评估期间减轻和检测偏差的策略及注意事项。分割、诊断、卷积神经网络（CNN）©RSNA，2022年
