• Opto-Electronic Engineering
  • Vol. 50, Issue 1, 220118 (2023)
Jingyu Liu1, Huaiyu Cai1、*, Wenyue Hao1, Tingtao Zuo2, Zhongwei Jia3, Yi Wang1, and Xiaodong Chen1
Author Affiliations
  • 1Key Laboratory of Optoelectronic Information Technology Ministry of Education, School of Precision Instrument & Opto-electronics Engineering, Tianjin University, Tianjin 300072, China
  • 2Lepu Medical Technology (Beijing) Co., Ltd., Beijing 102200, China
  • 3Southwestern Lu Hospital, Liaocheng, Shandong 252325, China
  • show less
    DOI: 10.12086/oee.2023.220118 Cite this Article
    Jingyu Liu, Huaiyu Cai, Wenyue Hao, Tingtao Zuo, Zhongwei Jia, Yi Wang, Xiaodong Chen. Intravascular ultrasound image segmentation combining polar coordinate modeling and a neural network[J]. Opto-Electronic Engineering, 2023, 50(1): 220118 Copy Citation Text show less
    Ideal hypothesis diagrams. (a) The mask image that meets the ideal hypothesis; (b) The situation that does not meet the ideal hypothesis
    Fig. 1. Ideal hypothesis diagrams. (a) The mask image that meets the ideal hypothesis; (b) The situation that does not meet the ideal hypothesis
    Modeling schematics. (a) Original image of IVUS; (b) Schematic diagram of modeling result. The intima contour and media contour are marked with red and green curves, respectively. The modeling results of the lumen area and plaque area are marked with red and green line segments, respectively
    Fig. 2. Modeling schematics. (a) Original image of IVUS; (b) Schematic diagram of modeling result. The intima contour and media contour are marked with red and green curves, respectively. The modeling results of the lumen area and plaque area are marked with red and green line segments, respectively
    The proposed dense distance of regression network
    Fig. 3. The proposed dense distance of regression network
    Schematic diagram of the intersection of the true value and the predicted value patch area. Note: For the convenience of observation, the true value ray and the predicted value ray are staggered by a certain angle, and the two are actually on the same ray
    Fig. 4. Schematic diagram of the intersection of the true value and the predicted value patch area. Note: For the convenience of observation, the true value ray and the predicted value ray are staggered by a certain angle, and the two are actually on the same ray
    The graph of JM changing with the number of rays
    Fig. 5. The graph of JM changing with the number of rays
    Visualization of segmentation results of different modeling methods
    Fig. 6. Visualization of segmentation results of different modeling methods
    Comparison of the visual effects of the segmentation results
    Fig. 7. Comparison of the visual effects of the segmentation results
    Linear regression analysis of key clinical parameters
    Fig. 8. Linear regression analysis of key clinical parameters
    Bland-Altman analysis of key clinical parameters
    Fig. 9. Bland-Altman analysis of key clinical parameters
    患者标号1234
    图像数量21839318168
    Table 1. Information of the IVUS dataset
    BackboneSEB numJMHD/mmPADTER
    MedLumPlaqueMedLumMedLum-
    ResNet1800.86300.85890.69350.23610.15010.10770.10380
    10.86580.85200.69470.22520.16030.10480.11740
    20.86590.86340.69790.21670.15550.10780.10390
    30.86550.85980.70160.22580.15130.11170.10380
    ResNet3400.88660.86740.73020.19060.14460.08490.09500
    10.88030.86060.71730.20800.14750.08910.10130
    20.87160.86100.70710.23570.15580.09790.10460
    30.88180.85740.71670.22270.15590.08330.10770
    ResNet5000.88040.86920.72210.18550.14620.09370.09670
    10.87380.86870.71310.22480.13850.09780.09460
    20.87600.86710.71520.22660.14450.09010.09900
    30.88050.87570.72110.21530.13510.09150.08790
    ResNet10100.88530.86650.72980.20140.13800.08600.09280
    10.89100.86460.73460.18730.15530.08160.10990
    20.89340.87380.74300.17610.13550.07940.10050
    30.89020.87250.73370.17750.13960.07450.08270
    ResNet15200.88520.86460.73020.20120.14120.08810.10360
    10.88450.87110.72560.20730.14020.08740.09580
    20.89740.87080.74240.18790.14810.07360.09620
    30.88490.86270.72550.21010.15450.08540.10880
    Table 2. The performance of the proposed method under different depths of backbone and different numbers of SEB modules
    Loss functionJMHD/mmPADTER
    MedLumPlaqueMedLumMedLum-
    Smoothl10.87320.86980.70470.21310.14360.09990.09300
    Ll+Lp0.88500.87570.73190.21450.13500.09120.09290
    Lm+Lp0.89040.86360.73130.19970.15090.07940.11160
    Ll+Lm0.88080.87360.71830.21720.14470.08940.08920
    IVUS Polar IoU Loss0.89340.87380.74300.17610.13550.07940.10050
    Table 3. Experimental results with different loss functions
    建模方式LossJMHD/mmPADTER
    MedLumPlaqueMedLumMedLum-
    EllipseSmoothl10.82080.81240.61000.26330.17930.13990.14020.0767
    PCM-PK0.87320.86980.70470.21310.14360.09990.09300
    Table 4. Experimental results with different modeling methods
    BackboneJMHD/mmPADTER
    MedLumPlaqueMedLumMedLum-
    SegNet[30]-0.88560.86180.71480.53671.57830.09480.12450.2328
    UNet[31]-0.88570.88460.73000.47000.17760.09560.09500.1319
    Deeplabv3+[29]ResNet1010.90260.88860.75670.24270.13020.06770.07870.0390
    OursResNet1010.89340.87380.74300.17610.13550.07940.10050
    Table 5. Performance comparison of different segmentation models
    斜率截距Pearson相关系数
    LCSA0.98250.23590.9427
    VCSA1.1259−1.39110.9626
    PCSA1.2016−1.50440.9432
    Table 6. Results of linear regression analysis of key clinical parameters
    均值均值偏移偏移程度/%离群值比例/%
    LCSA5.9898−0.1320−2.205.25
    VCSA15.8044−0.5628−3.566.59
    PCSA9.8146−0.4308−4.407.27
    Table 7. Results of Bland-Altman analysis of key clinical parameters
    Jingyu Liu, Huaiyu Cai, Wenyue Hao, Tingtao Zuo, Zhongwei Jia, Yi Wang, Xiaodong Chen. Intravascular ultrasound image segmentation combining polar coordinate modeling and a neural network[J]. Opto-Electronic Engineering, 2023, 50(1): 220118
    Download Citation