Machine Vision Based Conic Estimation: An Effective Tool for Reinforcement of Deep Learning in Measuring Fetal Head Circumference

The artificial intelligence is considered as an effective way to estimate the border around the fetal head basedon ultrasonic images. This approach has been remained as an open problem because of existing somechallenges such as low contrast of the image under various conditions and its low signal to noise ratio.Targets: In this article the machine vision based, conic estimation scheme is utilized as supplementaryprocessing of a VNET deep network in order to promote the performance of artificial intelligenece inmeasuring fetal head parameters. Firstly, a cost function is constructed by using a family of curves, each ofthem may be a candidate for the fetal border. Then the objective function is minimized by using optimizationframework in order to obtain the best fitted curve.Methodology: The testbed for evaluation of the proposed idea was prepared by using the software Pythonwith Keras on a computer with a NVIDIA ۲۰۸۰ TI GPU with ۳۲ GB RAM. The tests were performed onHC۱۸ challenge dataset provided by Heuvel and collaborators which includes ۹۹۹ ultrasound images whichhave been divided into three categories of training, validation and test subsets, with ratios of ۶۰%, ۱۵% and۲۵% among the total images respectively.The obtained results showed that proposed scheme may lead to significant promotion in measuring theparameters of fetal head. Therefore our method resulted in the accuracy, precision dice and jaccardparameters equal to ۹۷.۶%, ۹۴.۳%, ۹۵.۴% and ۹۱.۷% respectively. These results show at least ۳.۶%improvement over the same methods which did not use from geometric concept.Conclusions: Improving the estimation parameters show effectiveness of our idea based on reinforcement ofartificial intelligence tools in measuring fetal head circumference by considering the geometric nature of theproblem.

کلیدواژه ها:

Fetal head circumference ، Deep learning ، Conic fitting ، Geometric properties