A Hybrid Intelligent System for Intrapartum Risk Evaluation: Integrating AI-Based Feature Extraction with Fuzzy Logic for Operative Delivery Prediction

Background and Objective: Assessment of operative vaginal delivery feasibility remains partly subjective despite the availability of intrapartum ultrasound parameters. The study aims to provide an objective "Delivery Color Code" (Green, Yellow, Red) based on automated evaluation of ultrasound images to assist clinicians in predicting the feasibility of OVD.Methods: Single-centre prospective study conducted on women in the second stage of labor having an indication for operative delivery. Upon enrolment, transperineal ultrasound images were acquired on the axial plane to document three sonographic predictors of the outcome of OVD, namely the position of the fetal occiput, the head-perineum distance (HPD) and the midline angle (MLA). The methodology integrates three distinct automatic algorithms for the real-time extraction of such ultrasound parameters, which are processed through a five-layer Adaptive Neuro-Fuzzy Inference System (ANFIS) architecture that combines the learning capabilities of neural networks with the transparency of fuzzy logic aiming to predict the outcome of OVD as follows: green – high chance of successful and easy OVD; yellow – high chance of challenging OVD; red – high chance of failed or unfeasible instrumental delivery. The study analyzed a dataset of 265 patients (164 "Green" cases, 80 "Yellow" cases, and 21 “Red” cases) for system training, while validation was performed on a balanced cohort of 63 cases.Results: The model achieved a macro-average accuracy of 88.9% on the validation test set. The system demonstrated high discriminative reliability at the risk extremes, yielding a sensitivity of 0.905 for both "Success" (Green) and "Failure" (Red) outcomes. Notably, the "Red" risk category achieved a precision of 0.950 and a specificity of 0.976, effectively eliminating critical misclassifications between safe and high-risk scenarios. Errors were primarily confined to the "Borderline" (Yellow) transition zone, where the model maintained a conservative safety bias.Conclusions: The automated diagnostic pipeline offers a robust, Explainable AI (XAI) framework for decision making when delivery is to be expedited in the second stage of labor. By providing objective metrics for fetal station and position, the system may assist clinicians on how to perform fetal extraction in the second stage of labor aiming to reduce maternal-fetal morbidity associated with failed OVD.