HWHaijun Wang

Haijun Wang

*This is an algorithmically generated author record based on publicly available information.

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


Haijun Wang has a total of 16 co-authors. Here are the top co-authors by number of shared publications:

  • Qiang Lin

    School of Mathematics and Computer Science, Northwest Minzu University, Lanzhou, China

    4 shared publications
  • Zhengxing Man

    School of Mathematics and Computer Science, Northwest Minzu University, Lanzhou, China

    4 shared publications
  • Yongchun Cao

    School of Mathematics and Computer Science, Northwest Minzu University, Lanzhou, China

    3 shared publications
  • Mingyang Luo

    Key Laboratory of China’s Ethnic Languages and Information Technology of Ministry of Education

    2 shared publications
  • Chengcheng Han

    School of Mathematics and Computer Science, Northwest Minzu University; No. 1, Xibei Xincun Rd. Lanzhou city, Gansu Province, China 730030 Tel.: +86 - 13909405358

    1 shared publication
  • Chuangui Cao

    School of Mathematics and Computer Science, Northwest Minzu University, Lanzhou, China

    1 shared publication

Latest Publications

  • Open Access

    Semi-supervised segmentation of metastasis lesions in bone scan images

    by Qiang Lin, Runxia Gao, Mingyang Luo, Haijun Wang, Yongchun Cao, Zhengxing Man, Rong Wang

    To develop a deep image segmentation model that automatically identifies and delineates lesions of skeletal metastasis in bone scan images, facilitating clinical diagnosis of lung cancer–caused bone metastasis by nuclear medicine physicians. A semi-supervised segmentation model is proposed, comprising the feature extraction subtask and pixel classification subtask. During the feature extraction stage, cascaded layers which include the dilated residual convolution, inception connection, and feature aggregation learn the hierarchal representations of low-resolution bone scan images. During the pixel classification stage, each pixel is first classified into categories in a semi-supervised manner, and the boundary of pixels belonging to an individual lesion is then delineated using a closed curve. Experimental evaluation conducted on 2,280 augmented samples (112 original images) demonstrates that the proposed model performs well for automated segmentation of metastasis lesions, with a score of 0.692 for DSC if the model is trained using 37% of the labeled samples. The self-defined semi-supervised segmentation model can be utilized as an automated clinical tool to detect and delineate metastasis lesions in bone scan images, using only a few manually labeled image samples. Nuclear medicine physicians need only attend to those segmented lesions while ignoring the background when they diagnose bone metastasis using low-resolution images. More images of patients from multiple centers are typically needed to further improve the scalability and performance of the model via mitigating the impacts of variability in size, shape, and intensity of bone metastasis lesions.


  • Open Access

    Stenting for pulmonary artery stenosis resistant to balloon pulmonary angioplasty in chronic thrombo-embolic pulmonary hypertension: a bail out strategy

    by Yunshan Cao, Vikas Singh, Na Jiang, Rong Wei, Kaiyu Jiang, Haijun Wang

    A 73-year-old man with chronic thrombo-embolic pulmonary hypertension (CTEPH) was referred for treatment with balloon pulmonary angioplasty (BPA). After four BPA sessions, flow to the right most basal segmental pulmonary arteries was restored. Panel A/B shows the computed tomography pulmonary angiography (CTPA) images at baseline while Panel D/E represents CTPA 4 months after the fourth BPA session (white arrows). However, the lung perfusion 5 months after the last BPA had no significant improvement (circle in Panel F) when compared to the perfusion before BPA (circle in Panel C). The total effects were significantly limited by a severe proximal ring-like lesion in the right basal trunk (Panels D and E and Video 1). During the fifth treatment session, we performed serial dilation of this stenosis using 5.0 mm × 20 mm and 8.0 mm × 20 mm balloons (SterlingTM MonorailTM, Boston Scientific) at the right basal trunk. The lesion was recalcitrant, and there was no significant change in the pressure gradient across the lesion (40 mmHg) or the degree of stenosis after deflating the balloon. Then, we implanted a 9.0 mm × 19 mm stent (ExpressTM Vascular LD, Boston Scientific) across this lesion. The stent was well-expanded, and the pressure gradient dissipated with no complications. At a 5-month follow-up after stenting, the stent was patent (arrows in Panels D and H, Video 2), and lung perfusion was improved (circle in Panel I) when compared to that of pre-stent (circle in Panel F). Also, tricuspid annular plane systolic excursion increased from 13 mm at baseline to 17 mm about 4 months after the fourth BPA session to 21 mm 5 months after stenting; N-terminal pro-brain natriuretic peptide decreased from 6232 pg/mL at baseline to 2540 pg/mL about 4 months after the fourth BPA session to 814.6 pg/mL 5 months after stenting.


  • Open Access

    Deep learning based automated diagnosis of bone metastases with SPECT thoracic bone images

    by Qiang Lin, Tongtong Li, Chuangui Cao, Yongchun Cao, Zhengxing Man, Haijun Wang

    SPECT nuclear medicine imaging is widely used for treating, diagnosing, evaluating and preventing various serious diseases. The automated classification of medical images is becoming increasingly important in developing computer-aided diagnosis systems. Deep learning, particularly for the convolutional neural networks, has been widely applied to the classification of medical images. In order to reliably classify SPECT bone images for the automated diagnosis of metastasis on which the SPECT imaging solely focuses, in this paper, we present several deep classifiers based on the deep networks. Specifically, original SPECT images are cropped to extract the thoracic region, followed by a geometric transformation that contributes to augment the original data. We then construct deep classifiers based on the widely used deep networks including VGG, ResNet and DenseNet by fine-tuning their parameters and structures or self-defining new network structures. Experiments on a set of real-world SPECT bone images show that the proposed classifiers perform well in identifying bone metastasis with SPECT imaging. It achieves 0.9807, 0.9900, 0.9830, 0.9890, 0.9802 and 0.9933 for accuracy, precision, recall, specificity, F-1 score and AUC, respectively, on the test samples from the augmented dataset without normalization.


  • Open Access

    Automated Classification of Textual SPECT Diagnostic Reports with TextCNN Model

    by Chengcheng Han, Qiang Lin, Zhengxing Man, Yongchun Cao, Haijun Wang

    The SPECT diagnostic text contains several aspects of the patient’s personal information, image description, and suggested results. In order to construct a diagnostic model of nuclear medical text, a classification method of nuclear medical text based on deep learning was proposed. TextCNN was applied to propose the classification method of diseases. A set of real nuclear medical text data was used to verify the proposed method, the experimental results show that the proposed method has a good classification effect.


  • Automatic Segmentation of Metastasis Lesions in SPECT Bone Scan Images

    by Ruiting Gao, Qiang Lin, Mingyang Luo, Zhengxing Man, Yongchun Cao, Haijun Wang

    SPECT bone imaging is among the main nuclear medicine functional imaging modalities, which has the potential of early diagnosis of various serious diseases such as cancers. However, the low resolution and contrast of SPECT bone images bring huge challenges to performance on segmenting hotspots or lesions in SPECT bone scan images. In this work, three different algorithms including K-means clustering method, region growth method and C- V model are introduced to segment lesions in SPECT bone scan images by fine-tuning parameters, focusing on the bone metastatic lesions area in SPECT imaging. Specifically, each of the original 256 × 1024 whole body images is cropped into a 256 × 256 thoracic region including ribs and spine, followed by the data normalization. Then, different parameters are experimentally determined for the above three algorithms, leading to three different segmentation algorithms. Last, experimental evaluation conducted on a group of real-world samples of SPECT bone scan images reveals that our methods are workable for segmenting lesions with SPECT imaging, obtaining a value of 0.7307, 0.7768 and 0.8076 for Tanimoto similarity coefficient metric by the three methods, respectively. Particularly, the C- V model based segmentation method is able to provide more assistive information for oncologists on diagnose of tumors and other related diseases.