Automated ultrasound assessment of amniotic fluid index using deep learning

The estimation of antenatal amniotic fluid (AF) volume (AFV) is important as it offers crucial information about fetal development, well-being, and perinatal prognosis. However, AFV measurement cumbersome patient specific. Moreover, heavily sonographer-dependent, with accuracy varying greatly depending on the sonographer’s experience. Therefore, development accurate, robust, adoptable methods to evaluate highly desirable. In this regard, automation expected reduce user-based variability workload sonographers. automating very challenging, because accurate detection AF pockets difficult owing various confusing factors, such reverberation artifact, mimicking region floating matter. Furthermore, pocket exhibits an unspecified variety shapes sizes, ultrasound images often show missing or incomplete structural boundaries. To overcome abovementioned difficulties, we develop a hierarchical deep-learning-based method, which consider clinicians’ anatomical-knowledge-based approaches. key step segmentation using our proposed deep learning network, AF-net. AF-net variation U-net combined three complementary concepts - atrous convolution, multi-scale side-input layer, side-output layer. experimental results demonstrate that method provides index (AFI) robust precise from clinicians. achieved Dice similarity 0.877±0.086 for mean absolute error 2.666±2.986 relative 0.018±0.023 AFI value. best knowledge, first time, automated AFI.