V. Conclusion
In this work, a hybrid framework is designed for fetal head standard
plane detection. The proposed framework composites of two parts, i.e.,
object detection network and object classification network. With the
power of two networks, the designed framework is able to identify five
types of standard planes with satisfactory performance. By introducing
proposed model stacking, the performance of the proposed framework can
be further improved. Comparing with the state-of-the-art of fetal
ultrasound standard plane classification, i.e., SonoNet64, the average
accuracy has been boosted to 0.8961. The average AUC is 0.9893 which is
also indicated the effectiveness of the proposed hybrid framework. Since
the experiments are designed to reproduce the scenario happened in the
real life, the proposed method could be potentially applied to the
automatic fetal screening and diagnosis.