Application of a hybrid artificial intelligence model integrating view detection and structural segmentation in evaluating cardiac function of anemic fetuses
10.3760/cma.j.cn131148-20250228-00110
- VernacularTitle:基于切面检测与结构分割混合人工智能模型评估贫血胎儿心功能
- Author:
Yujun HUANG
1
;
Yunxiao ZHU
;
Kun YUAN
;
Nan WANG
;
Xiaomin ZHU
;
Qingying LI
;
Kangting WANG
;
Qun FANG
Author Information
1. 中山大学附属第七医院超声科，深圳 518107
- Publication Type:Journal Article
- Keywords:
Artificial intelligence;
Fetal anemia;
Echocardiography;
Fetal heart function;
Segmented quantitative analysis
- From:
Chinese Journal of Ultrasonography
2025;34(7):586-593
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To compare the cardiac size，morphology，and function between anemic and normal fetuses using a hybrid artificial intelligence（AI）model，and to evaluate the utility of AI in quantitatively assessing fetal cardiac function in cases of anemia.Methods:A retrospective study was conducted by collecting data from 2018 to 2024 at the Seventh Affiliated Hospital of Sun Yat-sen University，including 15 cases of anemic fetuses（anemia group）diagnosed through umbilical venous puncture and 32 cases of normal fetuses（control group）. Four-chamber fetal cardiac ultrasound videos and left/right ventricular segments were included，with 44 videos and 1 056 segments in the anemia group，and 46 videos and 1 104 segments in the control group. Based on dynamic four-chamber heart images，the hybrid AI model was employed to extract heart measurement parameters，including basal-apical length（BAL），transverse width（TW），global sphericity index（GSI），end-diastolic area（EDA），24-segment left and right ventricular end-diastolic diameter（LVEDD，RVEDD），segmental sphericity index（LVSI，RVSI），global longitudinal strain（LVGLS，RVGLS），fractional area change（LVFAC，RVFAC），segmental fractional shortening（LVFS，RVFS），along with their corresponding Z-scores. The differences in cardiac size，morphology，and function parameters between the two groups were compared. Pearson correlation analysis was performed for the parameters of the control group（BAL，TW，EDA，GLS，LVGLS，RVGLS，LVFAC，and RVFAC）against gestational age. The measurement consistencies of AI technology and fetal HQ technology in normal and anemia groups were evaluated.Results:No significant differences were found in BAL，TW，EDA，or GSI between groups（all P>0.05）. RVEDD in segments 3-24 was significantly larger in the anemia group（all P<0.05），with significantly higher Z-score abnormality rates for LVEDD and RVEDD across 24 segments（both P<0.001）. LVSI in segments 7-10，12，14-15 and RVSI in segments 1-23 were lower in the anemia group（all P<0.05），with significantly higher Z-score abnormality rates for LVSI and RVSI across 24 segments（both P<0.001）. The absolute values of LVGLS and LVFAC were significantly reduced in the anemia group（both P<0.05），while the absolute values of RVGLS and RVFAC showed no significant differences（both P>0.05）. Segmental LVFS values were significantly lower in the anemia group for segments 2，5-8，11-13（all P<0.05）. In the control group，BAL，TW，and EDA positively correlated with gestational age（ r=0.913，0.947，0.907；all P<0.001），while GSI，LVGLS，RVGLS，LVFAC and RVFAC showed no or weak correlations（ r=-0.221，0.353，0.515，-0.409，-0.425）. The intraclass correlation coefficient（ICC）between AI-based and conventional fetal HQ evaluations were 0.788 for the control group and 0.837 for the anemia group，indicating good consistency. Conclusions:AI offers a reliable approach for quantitatively evaluating fetal cardiac size，shape，and systolic function. Fetal anemia primarily affects right ventricular morphology and left ventricular systolic performance，characterized by spherical remodeling of the right ventricle and reductions in LVGLS，LVFAC，and segmental LVFS. The hybrid AI model holds potential value in fetal cardiac function assessment.
