分析评价一种新型语义网络分析模型SPReCHD：残差学习诊断系统模块（RLDS）和双路径链式多尺度门控轴心变压器网络模块（DPC-MSGATNet），在室间隔完整型肺动脉闭锁（PA/IVS）和危重肺动脉瓣狭窄（CPS）胎儿超声心动图四腔心视图数据集中评估右心发育不良程度分级的性能。

回顾性收集2017年6月至2022年12月青岛大学附属妇女儿童医院350例24~28周胎龄的PA/IVS和CPS胎儿的1650张单幅超声心动图四腔心视图建立实验数据集。根据右心发育不良程度对建立的SPReCHD模型进行训练、验证和测试，评估模型精确度、召回率及F1值等性能指标；与高级医师基于超声指标、右心室形态及出生结局等多维度信息对测试集做出的评估结果进行比较，采用Kappa检验观察评估的一致性。

SPReCHD模型在训练集、验证集及测试集对右心发育不良程度分级评估的精确度分别为93.82%、94.34%和94.68%，召回率分别为90.54%、91.38%和90.89%，F1值分别为92.10%、92.82%和92.67%。在测试集中，SPReCHD模型与高级儿童心血管医师评估结果一致性良好（Kappa值=0.724，P<0.001）。

SPReCHD模型在训练集、验证集及测试集对右心发育不良程度分级的评估性能优异。该模型对PA/IVS和CPS胎儿右心发育不良程度分级评估水平与高级医师评估结果一致性好，为进一步提出精准评估标准奠定基础。

To evaluate the performance of a new semantic parsing network (SPReCHD model: the multi-level residual mixed attention mechanism module [RLDS] and the dual path chain multi-scale gated axial transformer network module [DPC-MSGATNet]) in assessing the degree of right ventricular hypoplasia in fetuses with pulmonary atresia with intact ventricular septum (PA/IVS) or critical pulmonary stenosis (CPS) via fetal echocardiography four-chamber views.

This study retrospectively collected 1650 single-frame four-chamber views of 350 PA/IVS and CPS fetuses at 24 to 28 weeks gestation at Women and Children's Hospital, Qingdao University between June 2017 to December 2022 to establish an experimental dataset. The established SPReCHD model was trained, validated, and tested to grade right heart hypoplasia, with an evaluation of model accuracy, recall rate, and F1 score. The evaluation results of the SPReCHD model and senior physicians in the test set were compared based on multi-dimensional information such as ultrasound indicators and right ventricular morphology. Kappa test was used to observe the consistency of the evaluation results.

The SPReCHD model showed a precision of 93.82%, 94.34%, and 94.68%, recall rate of 90.54% , 91.38%, and 90.89%, and F1 score of 92.10%, 92.82%, and 92.67% in the training, validation, and test sets for evaluating the degree of right ventricular hypoplasia, respectively. In the test set, there was good consistency between the evaluation results of senior physicians and the SPReCHD model (Kappa=0.724, P<0.001).

The SPReCHD model has excellent performance in evaluating the degree of right ventricular hypoplasia in the training, validation, and test sets, achieving the expected level of research and development. There is a high level of concordance between the evaluation results for fetal right ventricular dysplasia by the model and senior physicians in fetuses with PA/IVS and CPS, thus laying a foundation for proposing precise evaluation standards.