Learning Tumor Growth via Follow-Up Volume Prediction for Lung - - PowerPoint PPT Presentation
Learning Tumor Growth via Follow-Up Volume Prediction for Lung Nodules Yamin Li 1 , Jiancheng Yang 1 , Yi Xu 1 , Jingwei Xu 1 , Xiaodan Ye 2 , Guangyu Tao 2 , Xueqian Xie 3 , Guixue Liu 3 1 Shanghai Jiao Tong University, 2 Shanghai Chest Hospital,
Shanghai Jiao Tong University
Yamin Li1, Jiancheng Yang1, Yi Xu1, Jingwei Xu1, Xiaodan Ye2, Guangyu Tao2, Xueqian Xie3, Guixue Liu3
1Shanghai Jiao Tong University, 2Shanghai Chest Hospital, 3Shanghai General Hospital
Ø Follow-up serves an important role in the management of pulmonary nodules for lung cancer. Ø Recent deep learning studies using convolutional neural networks (CNNs) to predict the malignancy score of nodules, only provides clinicians with black-box predictions and needs pathological labels. Ø Accurate prediction of tumor growth could help radiologists evaluate the risk of lung nodules without pathological examination and make clinical decision for each patient. l
Background: Help radiologists with lung nodule management
l
Task: To predict future volume of a lung nodule given any time interval and a baseline volume.
Ø Cases with long-term follow-up and LDCT scans are extremely difficult to obtain. Ø Currently we have more than 300 pulmonary nodules and more than 700 follow-up pairs collected from Shanghai Chest Hospital and Shanghai General Hospital. Ø Each nodule has a pixel-level label annotated by professional radiologists. Ø All nodules of one patient are divided into the same subset while performing 5-fold cross-validation.
Time point 1 Time point 2 Time point 3
l
Dataset:
Δ𝑢! Δ𝑢" Δ𝑢! + Δ𝑢" 3 pairs
Temporal Encoding:
PSNR* means PSNR inside the nodule