Advertisement

25: Performance of an automatic quantitative ultrasound analysis (AQUA) texture extractor to predict fetal lung maturity assessed by TDx-FLM in amniotic fluid

      Objective

      To evaluate the performance of a non-invasive Automatic Quantitative Ultrasound Analysis (AQUA software) texture extractor to predict fetal lung maturity in amniotic fluid as assessed by TDx-FLM.

      Study Design

      69 women with singleton pregnancies at 24.6-40.2 weeks' gestational age and undergoing amniocentesis to assess fetal lung maturity (TDx-FLM method) status. TDx-FLM result was categorized as mature or immature using standard normative values. An axial four-chamber fetal thorax image was acquired by ultrasound, and a fixed-box was placed in the fetal lung area. AQUA analyzed the pixels in the box and transformed them into a set of texture descriptors (>15,000). A Sequential Forward Selection technique with a Fishers objective function was applied to extract the most relevant imaging biomarkers, which were then input to a Support Vector Machines model able to learn from them and ultimately distinguish between a mature or immature status of the amniotic fluid by TDx-FLM. A leave one out method was employed in order to attain unbiased and realistic results.

      Results

      Mean (SD) of gestational age was 31.8 (4.7) weeks. According to TDx-FLM results, 22 samples of amniotic fluid demonstrated lung maturity and 47 did not. The imaging biomarker based on AQUA evaluation presented a sensitivity of 86%, a specificity of 98%, and an overall accuracy of 94% in detecting mature or immature status of the amniotic fluid by TDx-FLM.

      Conclusion

      Fetal lung image textures extracted by AQUA provided robust features to predict amniotic fluid TDx-FLM results. These results should be confirmed in larger sample sizes, which would allow much better predictive algorithms. This supports further research on quantitative imaging biomarkers of fetal lung maturity, which might avoid the need for amniocentesis in clinical settings.
      Figure thumbnail grr3