Xiaotong Gui
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Xiaotong Gui has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Adriel Saporta
Stanford University Department of Computer Science, USA
1 shared publication Andrew Y. Ng
Stanford University Department of Computer Science, USA
1 shared publication- Anuj Pareek
Department of Radiology, Copenhagen University Hospital, Copenhagen, Denmark
1 shared publication - Ashwin Agrawal
Department of Civil and Environmental Engineering Stanford University Stanford CA USA
1 shared publication Chanh Dt Nguyen
VinBrain, Vietnam
1 shared publicationFrancis G. Blankenberg
Stanford University School of Medicine, Department of Radiology, USA
1 shared publication
Latest Publications
- Open Access
Benchmarking saliency methods for chest X-ray interpretation
Saliency methods, which “explain” deep neural networks by producing heat maps that highlight the areas of the medical image that influence model prediction, are often presented to clinicians as an aid in diagnostic decision-making. Although many saliency methods have been proposed for medical imaging interpretation, rigorous investigation of the accuracy and reliability of these strategies is necessary before they are integrated into the clinical setting. In this work, we quantitatively evaluate seven saliency methods—including Grad-CAM, Grad-CAM++, and Integrated Gradients—across multiple neural network architectures using two evaluation metrics. We establish the first human benchmark for chest X-ray segmentation in a multilabel classification set up, and examine under what clinical conditions saliency maps might be more prone to failure in localizing important pathologies compared to a human expert benchmark. We find that (i) while Grad-CAM generally localized pathologies better than the other evaluated saliency methods, all seven performed significantly worse compared with the human benchmark; (ii) the gap in localization performance between Grad-CAM and the human benchmark was largest for pathologies that were smaller in size and had shapes that were more complex; (iii) model confidence was positively correlated with Grad-CAM localization performance. While it is difficult to know whether poor localization performance is attributable to the model or to the saliency method, our work demonstrates that several important limitations of saliency methods must be addressed before we can rely on them for deep learning explainability in medical imaging.
Open AccessBenchmarking saliency methods for chest X-ray interpretation
Saliency methods, which “explain” deep neural networks by producing heat maps that highlight the areas of the medical image that influence model prediction, are often presented to clinicians as an aid in diagnostic decision-making. Although many saliency methods have been proposed for medical imaging interpretation, rigorous investigation of the accuracy and reliability of these strategies is necessary before they are integrated into the clinical setting. In this work, we quantitatively evaluate seven saliency methods—including Grad-CAM, Grad-CAM++, and Integrated Gradients—across multiple neural network architectures using two evaluation metrics. We establish the first human benchmark for chest X-ray segmentation in a multilabel classification set up, and examine under what clinical conditions saliency maps might be more prone to failure in localizing important pathologies compared to a human expert benchmark. We find that (i) while Grad-CAM generally localized pathologies better than the other evaluated saliency methods, all seven performed significantly worse compared with the human benchmark; (ii) the gap in localization performance between Grad-CAM and the human benchmark was largest for pathologies that were smaller in size and had shapes that were more complex; (iii) model confidence was positively correlated with Grad-CAM localization performance. While it is difficult to know whether poor localization performance is attributable to the model or to the saliency method, our work demonstrates that several important limitations of saliency methods must be addressed before we can rely on them for deep learning explainability in medical imaging.
- 15 November 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Letters
- Vol. 3 (1) , 1-4
Multimodal Data Fusion in 3-D Printing Quality Prediction
Additive manufacturing, or three-dimensional (3D) printing, is an emerging technology that enables the direct fabrication of complex shapes. The material solidification in 3D printing, however, may yield geometric shape deviation in the 3D-printed parts, which adversely affects product quality and performance in engineering applications. The advancement in sensing technology enables us to collect measurement data to facilitate automated quality inspection for 3D printing. The objective of this study is to develop a data fusion method that can effectively fuse the different sources of sensor data and process data to evaluate the quality of 3D-printed parts. Specifically, this study fuses the information extracted from three data sources with different accuracy and measurement efficiency to evaluate the geometric shape deviation of the printed parts. Significant features about part quality are extracted from the 3D surface measurement data using Principal Component Analysis and Uncorrelated Multilinear Principal Component Analysis. Selected features are then fed into classification and regression to infer the quality of the part. The proposed prediction model starts with a preliminary classifier for the initial inspection using only low-resolution 3D surface measurement, followed by a more refined classifier trained by high-resolution 3D surface measurement to update the quality class labels. Such a two-level classification model is able to guarantee time efficiency while maintaining high accuracy. Results show that the proposed fusion method is promising for efficient and automated quality inspection.
- 15 November 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Letters
- Vol. 3 (1) , 1-4
Multimodal Data Fusion in 3-D Printing Quality Prediction
Additive manufacturing, or three-dimensional (3D) printing, is an emerging technology that enables the direct fabrication of complex shapes. The material solidification in 3D printing, however, may yield geometric shape deviation in the 3D-printed parts, which adversely affects product quality and performance in engineering applications. The advancement in sensing technology enables us to collect measurement data to facilitate automated quality inspection for 3D printing. The objective of this study is to develop a data fusion method that can effectively fuse the different sources of sensor data and process data to evaluate the quality of 3D-printed parts. Specifically, this study fuses the information extracted from three data sources with different accuracy and measurement efficiency to evaluate the geometric shape deviation of the printed parts. Significant features about part quality are extracted from the 3D surface measurement data using Principal Component Analysis and Uncorrelated Multilinear Principal Component Analysis. Selected features are then fed into classification and regression to infer the quality of the part. The proposed prediction model starts with a preliminary classifier for the initial inspection using only low-resolution 3D surface measurement, followed by a more refined classifier trained by high-resolution 3D surface measurement to update the quality class labels. Such a two-level classification model is able to guarantee time efficiency while maintaining high accuracy. Results show that the proposed fusion method is promising for efficient and automated quality inspection.