Evaluation Of Fetal Cerebral Cortical Development Based On Five Axial Planes Of The Brain Combined With Artificial Intelligence Technology

BackgroundMalformations of cortical development(MCD)can lead to diseases such as epilepsy,cognitive impairments,and sensory-motor developmental disorders.Prenatal diagnosis is the most effective measure to reduce the birth of children with such severe defects and prevent the economic losses to families and society.The development of fetal brain sulci follows a certain pattern,which can reflect the development of the cortical surface.Understanding the regular development of fetal brain sulci is the basis for prenatal screening of MCD.Prenatal diagnosis of MCD requires highly specialized knowledge and interdisciplinary collaboration.The general process is to first find suspicious signs by ultrasound,followed by further cranial MRI or genetic testing.Therefore,prenatal ultrasound is the first and most important step.Conventional cranial screening only scans the region from the posterior horn of the lateral ventricle to the cerebellum,and the cerebral cortex from the body of the lateral ventricle to the cranial vertex cannot be evaluated.Therefore,our research team has proposed a more comprehensive cerebral evaluation method called the five axial planes of the fetal brain.It adds the transcalvarial and the trans-genu-and-splenium axial plane to the existing three axial planes,which can screen for more severe brain abnormalities without significantly increasing the examination time.It can also provide a more comprehensive evaluation of cortical development.However,further research is needed to determine the specific ultrasound signs in prenatal ultrasound that indicate fetal MCD and guide further examinations.In recent years,artificial intelligence has been widely applied in the field of ultrasound.Our joint research team with Hunan University has made some progress in intelligent prenatal ultrasound.Therefore,combining the existing research on normal and abnormal morphology of the sylvian fissure with artificial intelligence can assist ultrasound physicians in evaluating cortical development.Based on the above considerations,this paper aims to gradually improve the prenatal ultrasound screening and clinical diagnostic strategies for fetal MCD,from establishing regular ultrasound evaluation methods and screening criteria to selecting the clinical process after discovering abnormal ultrasound signs.Additionally,it seeks to leverage the power of artificial intelligence for assisting in the diagnosis,thereby enhancing the focus and confidence in evaluating cortical development during examinations.Part 1 Study on the stability of primary sulci visualization based on five axial planes of the fetal brainObjectiveTo determine the optimal observation planes of primary sulci on five axial planes of the fetal brain between 20 and 32 weeks of gestation.Materials and methodsA total of 561 cases of normal singleton pregnancies between September 2019 and June 2022,who underwent ultrasound examinations at Shenzhen Maternity and Child Healthcare Hospital were included.The display of each primary sulcus on the five axial planes of the fetal brain was observed and recorded.Far-field sulci and sylvian fissure were scored as 0 for not displayed and 1 for displayed.If the sylvian fissure island formation occurred,the length of the platform was measured three times and the average value was calculated.The midline sulcus was scored as 2 for displayed on both sides,1 for displayed on one side,and 0 for not displayed on both sides.Fetal gestational age and pregnancy outcomes were recorded.Statistical analysis was performed using SPSS 25.0 software,including chi-square test,Mann-Whitney U test,kappa coefficient,and paired t-test.ResultsThe transcalvarial axial plane was the optimal plane for observing bilateral parieto-occipital sulcus,cingulate sulcus,and far-field central sulcus,precentral sulcus,postcentral sulcus and intraparietal sulcus.The transventricular axial plane was the optimal plane for observing the far-field inferior frontal sulcus,superior temporal sulcus,and inferior temporal sulcus.Among them,there was large individual variation in the development of the superior and inferior temporal sulci,suggesting dynamic scanning.The transthalamic axial plane had the greatest length of the sylvian fissure island platform and could comprehensively evaluate the development of the island platform and bilateral opercula,making it the optimal observation plane for the sylvian fissure.The superior frontal sulcus had low display rates on all five axial planes and generally poor repeatability,so its evaluation on five axial planes is not recommended.ConclusionThe five axial planes of the fetal brain obtained through two-dimensional abdominal ultrasound provide a comprehensive assessment of the fetal primary sulci.Combined with dynamic scanning,evaluating the development of each sulcus on the optimal display plane is crucial for screening fetal cortical development.Part 2 Feasibility study of simplified evaluation method for fetal parieto-occipital sulcus development on transcalvarial axial plane of fetal brainObjectiveTo validate the morphological changes of the parieto-occipital sulcus in the transcalvarial axial plane between 20 and 32 weeks of gestation,simplify the grading method for assessing fetal parieto-occipital sulcus development,and confirm its clinical feasibility.Materials and methodsA total of 550 cases of normal singleton pregnancies between 20 and 32 weeks of gestation,who underwent routine ultrasound examinations at Shenzhen Maternity and Child Healthcare Hospital from September 2019 to June 2022,were included.The morphological changes of the parieto-occipital sulcus in the transcalvarial axial plane were observed.Based on the developmental features of the sulcus,including angle formation,gradual closure,and curved growth,the development of the parieto-occipital sulcus was classified into six types:straight or shallow arc shape(Grade 0),blunt-angle V shape(Grade 1),acute-angle V shape(Grade 2),Y shape(Grade 3),I shape(Grade 4),and curved shape(Grade 5).The gestational age at examination and pregnancy outcomes were recorded.Statistical analysis was performed using SPSS 25.0 software,including box plots,Mann-Whitney U test,weighted kappa coefficient and curve regression analysis.ResultsGrading of the left parieto-occipital sulcus was obtained in 549 cases,and grading of the right parieto-occipital sulcus was obtained in 550 cases.The developmental grading of the parieto-occipital sulcus increased from Grade 0 to Grade 5 as the gestational age progressed.Grade 0 appeared between 20～22 weeks,grade 1 between 20～27 weeks,grade 2 between 24～29 weeks,grade 3 between 26～30 weeks,grade 4 between 29～32 weeks,and grade 5 between 31～32 weeks.The distribution of grading did not differ significantly between the left and right sides(p>0.05).Among the 549 fetuses,grading of the parieto-occipital sulcus was asymmetric in 43 cases(7.83%),with a difference of one grade between the two sides.Weighted kappa analysis showed strong consistency in the developmental grading of the left and right parieto-occipital sulcus,indicating a generally symmetrical development(p<0.001).There were strong interobserver and intraobserver agreement.ConclusionThe transcalvarial axial plane obtained through abdominal two-dimensional ultrasound allows for a preliminary and rapid evaluation of bilateral parieto-occipital sulcus.It enables early assessment of fetal cortical development and serves as an important basis for prenatal diagnosis of fetal cortical developmental abnormalities.Part 3 The value of quantitative scoring of prenatal ultrasound indicators related to malformations of cortical development in predicting fetal genetic abnormalitiesObjectiveTo construct a predictive model for genetic abnormalities related to fetal MCD based on five axial planes,using various ultrasound indicators.The model aims to provide a reference for deciding whether to perform invasive genetic testing in prenatal management.Materials and methodsA total of 13,514 pregnant women who underwent prenatal ultrasound and genetic testing at Shenzhen Maternity and Child Healthcare Hospital between January 1,2016,and June 30,2022,were included.Among them,94 cases with genetic abnormalities related to fetal MCD were assigned to the case group,and 7,494 cases with normal genetic testing results were assigned to the control group.Thirteen ultrasound features on the five axial planes were analyzed.SPSS 25.0 software was used to perform chi-square tests and single-factor analysis to identify ultrasound indicators related to genetic abnormalities associated with fetal MCD.Based on the positive likelihood ratio(+LR),scores were assigned to individual indicators.The diagnostic performance of the quantitative scoring system was evaluated using ROC curves and AUC,and the cutoff value was determined based on the Youden index.The sensitivity and specificity of the quantitative scoring system in predicting genetic abnormalities related to fetal MCD were analyzed.ResultsMann-Whitney U test results showed no statistically significant differences in age and gestational age distribution between the two groups(p>0.05).The scores for delayed development of the central sulcus,abnormal morphology of the sylvian fissure,and delayed development of the parieto-occipital sulcus were 6,5,and 4,respectively.Abnormal corpus callosum development,irregular lateral ventricle walls,abnormal cerebellar development,and midline shift(curvature)were assigned 3 points each.Delayed development of the sylvian fissure and increased abnormal cerebral sulcus were assigned 2 points each.Absence of septum pellucidum and lateral ventricular fusion,small head circumference,and intracranial abnormal ultrasonic echoes were assigned 1 point each.ROC curve analysis of the quantitative scoring system yielded an AUC of 0.992(95%CI:0.989-0.994),indicating good predictive ability.Using the maximum Youden index principle,a total score of 2 was determined as the cutoff value,with sensitivity,specificity,and accuracy of 95.74%,97.69%,and 97.71%,respectively,indicating a significantly increased risk of fetal genetic abnormalities when the total score is≥2.ConclusionThe scoring system and predictive model constructed from the 13 ultrasound indicators are simple and practical,with high sensitivity,specificity,and accuracy.They can serve as a reference for prenatal ultrasound screening of fetal MCD and subsequent management decisions.Part 4 Study on morphological segmentation and classification of fetal sylvian fissure based on deep learningObjectiveTo investigate the feasibility of using deep learning algorithms for segmentation and classification of fetal sylvian fissure on the transthalamic axial plane.Materials and methodsWe collected routine prenatal ultrasound examinations performed at Shenzhen Maternity and Child Healthcare Hospital between September 2019 and June 2022.A total of 563 singleton or twin pregnancies with gestational age between 20 and 32 weeks were included as the normal group.113 cases showed abnormal morphology or delayed development of the fetal sylvian fissure on prenatal ultrasound as the abnormal group,among them 94 cases were confirmed to have MCD-related abnormalities through genetic testing and 19 cases through fetal cranial MRI.Finally,1795 ultrasound images from the normal group and 418 ultrasound images from the abnormal group were meticulously annotated for the sylvian fissure.Among the normal group,1588 images were used for training and 207 images for testing.Among the abnormal group,353 images were used for training and 65 images for testing.We compared the segmentation and classification performance of six models:mask2former,SOLOV2,queryinst,maskrcnn,detectors_htc,and yolact.The evaluation metrics for classification mainly included accuracy,recall,precision,and F1-score,while mean average precision(mAP)and Dice coefficient were used for segmentation evaluation.ResultsIn terms of segmentation performance,SOLOV2 achieved the highest mAP and Dice coefficient,which were 0.756 and 0.809,respectively.The yolact and SOLOV2 models had relatively low frame processing times of 0.051s and 0.063s,respectively,meeting the real-time requirements in clinical settings.Regarding classification performance,the SOLOV2 model showed the best results,with accuracy,precision,and F1-score reaching 0.9853,0.9951,and 0.9903,respectively.For recall,the maskrcnn,maskformer,and yolact models performed the best,achieving a recall rate of 0.9952,while the recall rates of other models ranged from 0.9807 to 0.9903.ConclusionThe SOLOV2 model can successfully segment the sylvian fissure from the transthalamic axial plane,and classify it by the segmented sylvian fissure morphology.