KDPre_RNN: Knowledge Distilled Pretrained Recurrent Neural Network for Tubal Ectopic Pregnancy Prediction

Ectopic pregnancy is an important cause to illness and death on a global scale. Early detection and selection of the most appropriate therapy for the patient are vital in order to prevent any consequences. Ectopic pregnancies need treatment, which may be administered by surgical intervention, pharmaceutical methods, or a conservative approach, depending on several conditions. This paper presents a novel approach to tubal ectopic pregnancy prediction using a Knowledge Distilled Pretrained Recurrent Neural Network (KDPre_RNN). By incorporating the power of transfer learning, the model utilizes a pretrained RNN architecture, fine-tuned on large pregnancy-related datasets, to enhance predictive performance on tubal ectopic pregnancy cases. Knowledge distillation is employed to compress the complex pretrained model into a more lightweight yet efficient version, ensuring reduced computational costs without compromising predictive accuracy. The KDPre_RNN model processes clinical features such as patient history, diagnostic tests, and ultrasound findings, learning temporal relationships crucial for early ectopic pregnancy detection. The proposed KDPre_RNN is analyzed using various metrics and compared with few state of art methods and found that it achieves 99.54% of accuracy, 98.35% of precision, 97.56% of recall and 94.54% of F1-score.