• Optics and Precision Engineering
  • Vol. 29, Issue 9, 2278 (2021)
Sai LI1, Hao-jiang LI2, Li-zhi LIU2, Tian-qiao ZHANG1, and Hong-bo CHEN1
Author Affiliations
  • 1School of Life & Environmental Science, Guilin University of Electronic Technology, Guilin54004, China
  • 2Department of Radiology, Sun Yat-sen University Cancer Center, Guangzhou 510060, China);Correponding author, E-mail: hongbochen@163.com
  • show less
    DOI: 10.37188/OPE.20212909.2278 Cite this Article
    Sai LI, Hao-jiang LI, Li-zhi LIU, Tian-qiao ZHANG, Hong-bo CHEN. Automatic location of anatomical points in head MRI based on the scale attention hourglass network[J]. Optics and Precision Engineering, 2021, 29(9): 2278 Copy Citation Text show less
    Anatomical point location map. RIA/LIA(special region display of right/left internal acoustic pore), RAS/LAS(right/left ascending segment of the internal carotid artery in the posterior cavernous sinus)
    Fig. 1. Anatomical point location map. RIA/LIA(special region display of right/left internal acoustic pore), RAS/LAS(right/left ascending segment of the internal carotid artery in the posterior cavernous sinus)
    Automatic location of anatomical points
    Fig. 2. Automatic location of anatomical points
    Size range of the anatomical point
    Fig. 3. Size range of the anatomical point
    Statistical chart of anatomical point location accuracy(Black dotted line in the figure is the dividing line of whether the result is accurate or not, while the left side of the dotted line is the accurate result and the right side is the inaccurate result)
    Fig. 4. Statistical chart of anatomical point location accuracy(Black dotted line in the figure is the dividing line of whether the result is accurate or not, while the left side of the dotted line is the accurate result and the right side is the inaccurate result)
    Renderings of different network algorithms. Red is the doctor's marker point and green is the algorithm's detection result.
    Fig. 5. Renderings of different network algorithms. Red is the doctor's marker point and green is the algorithm's detection result.
    解剖点SAHN原始HNFPN
    准确率(%)

    平均距离

    (像素)

    准确率(%)

    平均距离

    (像素)

    准确率(%)

    平均距离

    (像素)

    RIA83.02.7±2.974.53.2±5.426.36.7±4.4
    LIA82.32.7±1.574.03.3±5.729.76.8±4.2
    RAS90.32.4±3.184.63.2±8.824.37.4±4.3
    LAS93.01.9±2.086.33.0±4.525.37.1±4.0
    Table 1. Comparison table of results of different network algorithms
    解剖点3阶4阶5阶6阶
    准确率(%)

    平均距离

    (像素)

    准确率(%)

    平均距离

    (像素)

    准确率(%)

    平均距离

    (像素)

    准确率(%)

    平均距离

    (像素)

    RIA80.72.7±2.983.02.7±2.980.02.8±3.080.02.8±3.0
    LIA81.72.8±1.680.72.8±1.578.03.0±1.682.32.7±1.5
    RAS87.32.5±3.089.02.7±3.189.72.4±3.190.32.4±3.1
    LAS90.72.1±2.191.72.2±2.193.01.9±2.091.02.1±2.1
    Table 2. Comparison of results of different network depths
    Sai LI, Hao-jiang LI, Li-zhi LIU, Tian-qiao ZHANG, Hong-bo CHEN. Automatic location of anatomical points in head MRI based on the scale attention hourglass network[J]. Optics and Precision Engineering, 2021, 29(9): 2278
    Download Citation