Tianjiao Liu

Tianjiao Liu

Auto-generated profile, based on publicly available information

Tianjiao Liu

Department of Electronic Engineering, Tsinghua University, Beijing, China

Research Focus & Metrics

Academic Career Length

10Years (2017-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 December 2019 in Medical Image Analysis
Verified
Journal Article
Closed access
TLTianjiao Liu
Tianjiao Liu
QGQianqian Guo
Qianqian Guo
CLChunfeng Lian
Chunfeng Lian
XRXuhua Ren
Xuhua Ren
SLShujun Liang
Shujun Liang
DSDinggang Shen
Dinggang Shen

23 February 2022 in IEEE Journal of Biomedical and Health Informatics
Verified
Journal Article
Closed access
JLJintao Lu
Jintao Lu
XOXi Ouyang
Xi Ouyang
XSXueda Shen
Xueda Shen
TLTianjiao Liu
Tianjiao Liu
ZCZhiming Cui
Zhiming Cui
DSDinggang Shen
Dinggang Shen
Early detection and identification of malignant thyroid nodules, a vital precursory to the treatment, is a difficult task even for experienced clinicians. Many Computer-Aided Diagnose (CAD) systems have been developed to assist clinicians in performing this task on ultrasonic images. Learning-based CAD systems for thyroid nodules generally accommodate both nodule detection/ segmentation and fine-grained classification for its malignancy, and prior researches often treat aforementioned tasks in separate stages, leading to additional computational costs. In this paper, we utilize an online class activation mapping (CAM) mechanism to guide the network to learn discriminative features for identifying thyroid nodules in ultrasound images, called CAM attention network . It takes nodule masks as localization cues for direct spatial attention of the classification module, thereby avoiding isolated training for classification. Meanwhile, we propose a deformable convolution module to add offsets to the regular grid sampling locations in the standard convolution, guiding the network to capture more discriminative features of nodule areas. Furthermore, we use a generative adversarial network (GAN)to ensure reliable deformations of nodules from the deformable convolution module. Our proposed CAM attention network has already achieved the 2nd place in the classification task of TN-SCUI 2020, a MICCAI 2020 Challenge with the largest set of thyroid nodule ultrasound images according to our knowledge. The further inclusion of our proposed GAN-guided deformable module allows for capturing more fine-grained features between benign and malignant nodules, and further improves the classification accuracy to a new state-of-the-art level.

1 April 2017
Verified
Conference Paper
Closed access
TLTianjiao Liu
Tianjiao Liu
SXShuaining Xie
Shuaining Xie
YZYukang Zhang
Yukang Zhang
JYJing Yu
Jing Yu
LNLijuan Niu
Lijuan Niu
WSWeidong Sun
Weidong Sun
Ultrasonography is a valuable diagnosis method for thyroid nodules. Automatically discriminating benign and malignant nodules in the ultrasound images can provide aided diagnosis suggestions, or increase the diagnosis accuracy when lack of experts. The core problem in this issue is how to capture appropriate features for this specific task. Here, we propose a feature extraction method for ultrasound images based on the convolution neural networks (CNNs), try to introduce more meaningful and specific features to the classification. A CNN model trained with ImageNet data is transferred to the ultrasound image domain, to generate semantic deep features under small sample condition. Then, we combine those deep features with conventional features such as Histogram of Oriented Gradient (HOG) and Scale Invariant Feature Transform (SIFT) together to form a hybrid feature space. Furthermore, to make the general deep features more pertinent to our problem, a feature subset selection process is employed for the hybrid nodule classification, followed by a detailed discussion on the influence of feature number and feature composition method. Experimental results on 1037 images show that the accuracy of our proposed method is 0.929, which outperforms other relative methods by over 10%.