Explaining the Rationale of Deep Learning Glaucoma Decisions with Adversarial Examples
Explaining the Rationale of Deep Learning Glaucoma Decisions with Adversarial Examples
复制标题
DOI：
10.1016/j.opatha.2020.06.036
复制
发表时间：
2021-01-01
期刊：
影响因子：
13.7
通讯作者：
Park, Sang Min
中科院分区：
文献类型：
--
作者：
Chang, Jooyoung;Lee, Jinho;Park, Sang Min
目的：为了说明所谓的深度学习模型（DLM）黑匣子内部的内容，以便临床医生可以通过评估对抗性解释解释青光眼和青光眼相关发现的DLM决策的合理性的能力，对人工智能的结论有更大的信心。对抗性解释产生对抗性实例（AE），或已经改变以获得或失去病理特征特异性特征的图像，以解释DLM的基本原理。设计：DLM解释方法的评价。参与者：健康筛查参与者（n = 1653）在韩国首尔的首尔国立大学医院健康促进中心。我们使用6430张视网膜眼底图像对DLM进行了可参考性青光眼（RG）、杯盘比增加（ICDR）、椎间盘边缘狭窄（DRN）和视网膜神经纤维层缺损（RNFLD）的训练。调查包括使用AE和梯度加权类激活映射（GradCAM）的解释，一种传统的基于热图的解释方法，为400例病理和健康患者的眼睛。对于每种方法，经过董事会培训的青光眼专家对位置可解释性（在图像中精确定位决策相关区域的能力）和合理解释性（告知用户模型基于病理特征进行决策的推理的能力）进行了评级。评分采用配对Wilcoxon符号秩检验进行比较。主要结果测量：受试者工作特征曲线下面积（AUC）、DLMs的敏感性和特异性; AE的临床病理变化的可视化;以及位置和合理解释性的调查评分。RG、ICDR、DRN和RNFLD DLM的AUC分别为0.90、0.99、0.95和0.79，灵敏度分别为0.79、1.00、0.82和0.55，特异性为0.90。生成的AE显示了有效的临床特征变化，使用AE和GradCAM的位置可解释性调查结果分别为3.94 1.33和2.55 1.24，最高评分为5分。AE和GradCAM的合理解释性评分分别为3.97 ± 1.31和2.10 ± 1.25。对抗性的例子提供了显着更好的可解释性比GradCAM.Conclusions：对抗性的解释增加了GradCAM，传统的热图为基础的解释方法的可解释性。对抗性解释可以帮助医疗专业人员更清楚地理解DLMs用于临床决策时的基本原理。(C)2020年美国眼科学会
Purpose: To illustrate what is inside the so-called black box of deep learning models (DLMs) so that clinicians can have greater confidence in the conclusions of artificial intelligence by evaluating adversarial explanation on its ability to explain the rationale of DLM decisions for glaucoma and glaucoma-related findings. Adversarial explanation generates adversarial examples (AEs), or images that have been changed to gain or lose pathologic characteristic-specific traits, to explain the DLM's rationale.Design: Evaluation of explanation methods for DLMs.Participants: Health screening participants (n = 1653) at the Seoul National University Hospital Health Promotion Center, Seoul, Republic of Korea.Methods: We trained DLMs for referable glaucoma (RG), increased cup-to-disc ratio (ICDR), disc rim narrowing (DRN), and retinal nerve fiber layer defect (RNFLD) using 6430 retinal fundus images. Surveys consisting of explanations using AE and gradient-weighted class activation mapping (GradCAM), a conventional heatmapbased explanation method, were generated for 400 pathologic and healthy patient eyes. For each method, board-trained glaucoma specialists rated location explainability, the ability to pinpoint decision-relevant areas in the image, and rationale explainability, the ability to inform the user on the model's reasoning for the decision based on pathologic features. Scores were compared by paired Wilcoxon signed-rank test.Main Outcome Measures: Area under the receiver operating characteristic curve (AUC), sensitivities, and specificities of DLMs; visualization of clinical pathologic changes of AEs; and survey scores for locational and rationale explainability.Results: The AUCs were 0.90, 0.99, 0.95, and 0.79 and sensitivities were 0.79, 1.00, 0.82, and 0.55 at 0.90 specificity for RG, ICDR, DRN, and RNFLD DLMs, respectively. Generated AEs showed valid clinical feature changes, and survey results for location explainability were 3.94 1.33 and 2.55 1.24 using AEs and GradCAMs, respectively, of a possible maximum score of 5 points. The scores for rationale explainability were 3.97 1.31 and 2.10 1.25 for AEs and GradCAM, respectively. Adversarial example provided significantly better explainability than GradCAM.Conclusions: Adversarial explanation increased the explainability over GradCAM, a conventional heatmapbased explanation method. Adversarial explanation may help medical professionals understand more clearly the rationale of DLMs when using them for clinical decisions. (C) 2020 by the American Academy of Ophthalmology