Recognition of facial expression of fetuses by artificial intelligence (AI)

OBJECTIVES

The development of the artificial intelligence (AI) classifier to recognize fetal facial expressions that are considered as being related to the brain development of fetuses as a retrospective, non-interventional pilot study.

METHODS

Images of fetal faces with sonography obtained from outpatient pregnant women with a singleton fetus were enrolled in routine conventional practice from 19 to 38 weeks of gestation from January 1, 2020, to September 30, 2020, with completely de-identified data. The images were classified into seven categories, such as eye blinking, mouthing, face without any expression, scowling, smiling, tongue expulsion, and yawning. The category in which the number of fetuses was less than 10 was eliminated before preparation. Next, we created a deep learning AI classifier with the data. Statistical values such as accuracy for the test dataset and the AI confidence score profiles for each category per image for all data were obtained.

RESULTS

The number of fetuses/images in the rated categories were 14/147, 23/302, 33/320, 8/55, and 10/72 for eye blinking, mouthing, face without any expression, scowling, and yawning, respectively. The accuracy of the AI fetal facial expression for the entire test data set was 0.985. The accuracy/sensitivity/specificity values were 0.996/0.993/1.000, 0.992/0.986/1.000, 0.985/1.000/0.979, 0.996/0.888/1.000, and 1.000/1.000/1.000 for the eye blinking, mouthing, face without any expression, scowling categories, and yawning, respectively.

CONCLUSIONS

The AI classifier has the potential to objectively classify fetal facial expressions. AI can advance fetal brain development research using ultrasound.

Success rate of five fetal cardiac views using HDlive Flow with spatiotemporal image correlation at 18-21 and 28-31 weeks of gestation

Objective To assess the success rates of five fetal cardiac views using HDlive Flow (Silhouette) with spatiotemporal image correlation (STIC) in the second and third trimesters of pregnancy, and to verify the feasibility of obtaining five cardiac views by volumes. Methods One hundred and eighteen normal fetuses at 18-21 and 28-31 weeks of gestation were studied using HDlive Flow (Silhouette) with STIC to assess the success rates of five fetal cardiac views (frontal, spatial three-vessel, panoramic, posterior, and right lateral views). Target structures in each cardiac view were evaluated. "Good" was assigned when all structures were noted, "Fair" when only one structure was missed, and "Poor" when two and more structures could not been detected. Results There were no significant differences in success rates of each cardiac view between 18-21 and 28-31 weeks of gestation. The rate of "Good" with a spatial three-vessel view was significantly higher than that with other cardiac views at 18-21 and 28-31 weeks, respectively (P < 0.05). Conclusion Five cardiac views using HDlive Flow (Silhouette) with STIC may become an adjunctive and useful tool in fetal cardiac examination.

Evaluation of 24-segment sphericity index of fetal heart using FetalHQ

OBJECTIVE

To evaluate the 24-segment sphericity index (SI) of the fetal heart using FetalHQ regarding the change with advancing gestation and reproducibility.

METHODS

Eighty-one pregnant women at 18-21 + 6 and 28-31 + 6 weeks were studied using FetalHQ to calculate 24-segment SI of left and right ventricles. Intra- and inter-class correlation coefficients and intra- and inter-observer agreements of measurements for SI were assessed in each segment.

RESULTS

With respect to intra-observer reproducibility, all SI values showed correlations between 0.596 and 0.774 for the left ventricle, and between 0.491 and 0.764 for the right ventricle, with good intra-observer agreements except for right ventricular segment 11-15. With respect to inter-observer reproducibility, all SI values showed correlations between 0.473 and 0.755 for the left ventricle, and between 0.48 and 0.767 for the right ventricular segment 1-18. However, right ventricular segment 19-24 showed no correlations. Good inter-observer agreements were noted in the right ventricle, whereas poor inter-observer agreements were evident in the left ventricle. There were significant differences in the mean SI values in the apical segment of the left ventricle between 18-21 + 6 and 28-31 + 6 weeks (p < .05), whereas there were no significant differences in the mean SI values in the right ventricle between both gestational ages.

CONCLUSION

These results show that measurements of 24-segment SI of fetal heart had relatively low reliability of measurements performed by experienced sonographers. This may suggest that more training on proper placement of the 3 end-systolic and end-diastolic reference points during the initial analysis may be relevant.

A 24-segment fractional shortening of the fetal heart using FetalHQ

OBJECTIVES

To evaluate 24-segment fractional shortening (FS) of the fetal heart using FetalHQ by speckle-tracking regarding reproducibility and the change with advancing gestation.

METHODS

Eighty-one pregnant women at 18-21+6 and 28-31+6 weeks of gestation were studied using FetalHQ with the speckle-tracking technique to calculate 24-segment FS of left and right ventricles. Intra- and inter-class correlation coefficients and intra- and inter-observer agreements of measurements for FS were assessed in each segment.

RESULTS

With respect to intra-observer reproducibility, all FS values showed correlations between 0.575 and 0.862 for the left ventricle, with good intra-observer agreements except for left ventricular segments 14-24. Right ventricular FS values showed correlations between 0.334 and 0.685, with good intra-observer agreements. With respect to inter-observer reproducibility, all FS values showed correlations between 0.491 and 0.801 for the left ventricle, with good intra-observer agreements except for left ventricular segments 16-22. Right ventricular FS values showed correlations between 0.375 and 0.575, with good inter-observer agreements. There were significant differences in the mean FS values in the basal segment (segments 1-5) of the left ventricle between 18 and 21+6 and 28-31+6 weeks of gestation (p<0.05), whereas there were significant differences in all mean FS values in the right ventricle between both gestational ages (p<0.05).

CONCLUSIONS

These results suggest that the reproducibility of the 24-segment FS of the fetal heart using FetalHQ is fair. However, there may be significant differences in FS values with advancing gestational age, especially for the right ventricle.

Natural Course of Fetal Hyaloid Artery: SlowflowHD Longitudinal Study

OBJECTIVE

To assess the natural course of the fetal hyaloid artery (FHA) using SlowflowHD during pregnancy.

METHODS

One-hundred and eighteen normal fetuses were studied longitudinally using SlowflowHD to assess the natural course of FHA at 18-21 + 6, 28-31 + 6, and after 35 weeks of gestation. Blood flow of FHA with/without its branches (BF) was evaluated using SlowflowHD, and the band of FHA without blood flow (Band) was identified in each gestational period. The natural course of FHA was classified into six types: Type A, not detected (ND) in each gestational period; Type B, Band at 18-21 + 6 weeks; Type C, Band at 18-21 + 6 and 28-31 + 6 weeks; Type D, BF at 18-21 + 6 weeks; Type E, BF at 18-21 + 6 weeks and Band at 28-31 + 6 weeks; Type F, BF at 18-21 + 6 and 28-31 + 6 weeks.

RESULTS

BF, Band, and ND were 83.5, 15.5, and 1.0% at 18-21 + 6 weeks, 8.8, 15.5, and 75.7% at 28-31 + 6 weeks, and 0, 0, and 100% after 35 weeks, respectively. Type A was 1.0%, Type B was 13.6%, Type C was 1.9%, Type D was 61.2%, Type E was 13.6%, and Type F was 8.7%.

CONCLUSIONS

The presence of FHA blood flow before 30 weeks of gestation should be regarded as a physiologic phenomenon in normal fetuses.

Isolated small ventricular septal defects in utero: HDlive Flow with spatiotemporal image correlation and HDlive Silhouette case study

Purpose

To demonstrate shunt flow through small ventricular septal defects (VSDs) using HDlive Flow with spatiotemporal image correlation (STIC) and HDlive Silhouette.

Methods

Four cases of isolated small VSDs were evaluated using colour Doppler, and HDlive Flow with STIC and HDlive Silhouette.

Results

HDlive Flow with STIC and HDlive Silhouette can improve the ability to depict shunt flow through small VSDs compared with colour Doppler.

Conclusion

This technique provides information on spatial recognition of the shunt flow through isolated small VSDs in real time.

Artificial intelligence to understand fluctuation of fetal brain activity by recognizing facial expressions

OBJECTIVE

To examine whether artificial intelligence can achieve discoveries regarding fetal brain activity.

METHODS

In this observational study, the authors collected images of fetal faces using a four-dimensional ultrasound technique obtained from singleton pregnancies of outpatients in routine practice at 27 to 37 weeks of gestation between February 1 and December 31, 2021. The authors developed an artificial intelligence classifier to recognize seven facial expressions of fetuses, then applied it to video files of fetal facial images to generate the probabilities, as confidence scores, of each expression category. Discrete Fourier transform and chaotic analysis were used to investigate the scores. Mann-Whitney test, t test, variance test, and one-way analysis of variance were used for statistical analysis.

RESULTS

Facial expression changes were observed in cycles averaging 66 to 73 s. The power spectrum showed that mouthing and neutral expressions were the most prevalent. There was a difference between categories for the spectrum (p = 0.004). Two different states--dense and sparse--of confidence scores were discovered. The correlation dimension was 1.19 ± 0.22 and 1.33 ± 0.27 for dense and sparse, respectively (p = 0.047).

CONCLUSION

This method objectively and quantitatively demonstrated fetal brain activity and may provide insight into how the fetus spends its time in utero.

Isolated tortuous ductus arteriosus in a fetus: HDlive Flow with spatiotemporal image correlation (STIC) study

OBJECTIVES

To describe prevalence, outcome, color Doppler, and HDlive Flow (Silhouette) with spatiotemporal image correlation (STIC) features of isolated tortuous ductus arteriosus in Japanese fetuses at 28-31 + 6 weeks of gestation.

METHODS

During a 37-month period, 3,150 fetal screenings were performed at 28-31 + 6 weeks of gestation. Isolated tortuous ductus arteriosus was evaluated using color Doppler and HDlive Flow (Silhouette) with STIC. Prevalence and outcomes were also assessed.

RESULTS

There were 22/3,150 fetuses (0.7%) had isolated tortuous ductus arteriosus at 28-31 + 6 weeks of gestation. According to color Doppler and HDlive Flow findings, two types of isolated tortuous ductus arteriosus (Right-angled and Loop shapes) were classified. There were 20 Right-angled and 2 Loop shapes. One fetus with a Right-angled shape was associated with aortic tortuosity. HDlive Flow with STIC revealed spatial relationships of tortuous ductus arteriosus, aortic arch, and descending aorta. All fetuses with isolated tortuous ductus arteriosus had good neonatal outcomes.

CONCLUSIONS

The incidence of isolated tortuous ductus arteriosus in Japanese fetuses was low compared with those in previous reports. HDlive Flow with STIC provides precise information for spatial recognition and differentiation of isolated tortuous ductus arteriosus, resulting in two type classifications. Isolated tortuous ductus arteriosus in a fetus may be a transient, benign finding in utero.