Ultrasound image evaluation for nuchal translucency based on image recognition / 中华超声影像学杂志

ABSTRACT

Objective To establish the model and software for quality assessment of fetal nuchal translucency ultrasound image using computer image recognition technology.Methods The proposed approach firstly divided the input image into four sub-image blocksthe nasal bone(NB) area,the nuchal translucency (NT) area,the midbrain area,and the jaw and chest area.For each sub-image block,the algorithm compared the image block with the corresponding area of the standard training image set,and then determined whether the current image block was the qualified one using the the Gabor feature and Bayesian decision.The input ultrasound image was determined to be qualified only if it had four qualified sub-image blocks.Results The difference between our automatic method and the manual screening by experts wasting small,the method obtained Kappa =0.795 and P ＜0.001.Moreover,the efficiency of our method was much higher than the manual screening method.Conclusions Image recognition technology can effectively assist the sonographer to assess the quality of fetal NT of ultrasound image.The proposed approach can reduce the subjectivity and randomness of the manual evaluation of NT image.