Abstract
In this project, our goal is to develop a method for interpreting how a neural network makes layer-by-layer embedded decisions when trained for a classification task, and also to use this insight for improving the model performance. To do this, we first approximate the distribution of the image representations in these embeddings using random forest models, the output of which, termed embedding outputs, are used for measuring how the network classifies each sample. Next, we design a pipeline to use this layer embedding output to calibrate the original model output for improved probability calibration and classification. We apply this two-steps method in a fully convolutional neural network trained for a liver tissue classification task on our institutional dataset that contains 20 3D multi-parameter MR images for patients with hepatocellular carcinoma, as well as on a public dataset with 131 3D CT images. The results show that our method is not only able to provide visualizations that are easy to interpret, but that the embedded decision-based information is also useful for improving model performance in terms of probability calibration and classification, achieving the best performance compared to other baseline methods. Moreover, this method is computationally efficient, easy to implement, and robust to hyper-parameters.
Highlights
AND BACKGROUNDH EPATOCELLULAR carcinoma (HCC) is one of the most common malignancies in the liver, accounting for approximately 75 percent of all liver cancers, and is the third most common cause of cancer-related death worldwide [1]
Our main dataset consists of 20 sets of 3D multi-parameter magnetic resonance (MR) images on patients with HCC, each of which consisted of three T1 weighted dynamic contrast enhanced (DCE) images at three different time points: pre-contrast phase, arterial phase (20 seconds after the injection), and venous phase (70 seconds after the injection)
In the task of liver tissue classification in our institutional MRI dataset, we can use this to observe how the network extracts and refines the segmentation boundary layer by layer, and to trace the decision progression toward its final prediction, especially when it comes to a false positive anomaly prediction
Summary
H EPATOCELLULAR carcinoma (HCC) is one of the most common malignancies in the liver, accounting for approximately 75 percent of all liver cancers, and is the third most common cause of cancer-related death worldwide [1] Upon diagnosis, it is frequently treated with image-guided. Date of publication April 29, 2020; date of current version October 28, 2020. Color versions of one or more of the figures in this article are available online at http://ieeexplore.ieee.org. Such therapies, including the catheterbased transarterial chemoembolization (TACE), are known to have variable efficacy depending on tumor characteristics [2]. Advanced imaging, such as multi-parameter dynamic contrast enhanced (DCE) magnetic resonance (MR) imaging, is useful in quantifying such characteristics, evaluating a patient’s amenability to such therapies, and estimating tumor response [3]
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have