Oliveira Marta, Wilming Rick, Clark Benedict, Budding Céline, Eitel Fabian, Ritter Kerstin, Haufe Stefan
Division 8.44, Physikalisch-Technische Bundesanstalt, Berlin, Germany.
Computer Science Department, Technische Universität Berlin, Berlin, Germany.
Front Artif Intell. 2024 Feb 26;7:1330919. doi: 10.3389/frai.2024.1330919. eCollection 2024.
Convolutional Neural Networks (CNNs) are frequently and successfully used in medical prediction tasks. They are often used in combination with transfer learning, leading to improved performance when training data for the task are scarce. The resulting models are highly complex and typically do not provide any insight into their predictive mechanisms, motivating the field of "explainable" artificial intelligence (XAI). However, previous studies have rarely quantitatively evaluated the "explanation performance" of XAI methods against ground-truth data, and transfer learning and its influence on objective measures of explanation performance has not been investigated. Here, we propose a benchmark dataset that allows for quantifying explanation performance in a realistic magnetic resonance imaging (MRI) classification task. We employ this benchmark to understand the influence of transfer learning on the quality of explanations. Experimental results show that popular XAI methods applied to the same underlying model differ vastly in performance, even when considering only correctly classified examples. We further observe that explanation performance strongly depends on the task used for pre-training and the number of CNN layers pre-trained. These results hold after correcting for a substantial correlation between explanation and classification performance.
卷积神经网络（CNN）在医学预测任务中被频繁且成功地使用。它们经常与迁移学习结合使用，当该任务的训练数据稀缺时，能提高性能。由此产生的模型高度复杂，通常无法对其预测机制提供任何见解，这推动了“可解释”人工智能（XAI）领域的发展。然而，以往的研究很少针对真实数据定量评估XAI方法的“解释性能”，并且尚未研究迁移学习及其对解释性能客观指标的影响。在此，我们提出了一个基准数据集，该数据集能够在实际的磁共振成像（MRI）分类任务中量化解释性能。我们利用这个基准来了解迁移学习对解释质量的影响。实验结果表明，即使仅考虑正确分类的示例，应用于相同基础模型的流行XAI方法在性能上也存在巨大差异。我们进一步观察到，解释性能强烈依赖于用于预训练的任务以及预训练的CNN层数。在纠正了解释与分类性能之间的显著相关性之后，这些结果依然成立。
