Hybrid-FHR: a multi-modal AI approach for automated fetal acidosis diagnosis

• DOI: 10.1186/s12911-024-02423-4
• Publication Date: 2024-01-22T04:37:39Z
• Authors: Zhidong Zhao; Jiawei Zhu; Pengfei Jiao; Jinpeng Wang; Xiaohong Zhang; Xinmiao Lu; Yefei Zhang
• Keywords: Cardiotocography; Feature (linguistics)

Abstract Background In clinical medicine, fetal heart rate (FHR) monitoring using cardiotocography (CTG) is one of the most commonly used methods for assessing acidosis. However, as visual interpretation CTG depends on subjective judgment clinician, this has led to high inter-observer and intra-observer variability, making it necessary introduce automated diagnostic techniques. Methods study, we propose a computer-aided algorithm (Hybrid-FHR) acidosis assist physicians in objective decisions taking timely interventions. Hybrid-FHR uses multi-modal features, including one-dimensional FHR signals three types expert features designed based prior knowledge (morphological time domain, frequency nonlinear). To extract spatiotemporal feature representation signals, multi-scale squeeze excitation temporal convolutional network (SE-TCN) backbone model dilated causal convolution, which can effectively capture long-term dependence by expanding receptive field each layer’s convolution kernel while maintaining relatively small parameter size. addition, proposed cross-modal fusion (CMFF) method that multi-head attention mechanisms explore relationships between different modalities, obtaining more informative representations improving accuracy. Results Our ablation experiments show outperforms traditional previous methods, with average accuracy, specificity, sensitivity, precision, F1 score 96.8, 97.5, 96, 96.7%, respectively. Conclusions enables analysis, assisting healthcare professionals early identification prompt implementation