Three dimensional (3D) ultrasound and magnetic resonance imaging (MRI) – the future tools to diagnose pregnancies with high risk of adverse outcome?

Placental volume measurements using tracked 2D ultrasound and artificial intelligenceThe PhD-project of Karianne Sagberg concerns new ultrasound technology for the measuremets of placental size in ongoing pregnancies.The following manuscript is submitted for publication i 2023: Title: Placental volume measurements using tracked 2D ultrasound and artificial intelligence Authors: Sagberg, Karianne ,2, Torgrim Lie3, Peterson, Helene1,2, Hillestad, Vigdis2,4, Eskild, Anne1,2 Bø, Lars Eirik3 1Institute of Clinical Medicine, University of Oslo, Oslo, Norway 2Department of Obstetrics and Gynecology, Akershus University Hospital, Lørenskog, Norway 3Department of Health Research, SINTEF Digital, Trondheim, Norway 4Department of Diagnostic Imaging, Akershus University Hospital, Lørenskog, Norway Introduction/Purpose: Placental volume measurements can potentially help identify high-risk pregnancies. However, it is difficult to capture the entire placenta in a single 3D ultrasound image. We aimed to develop a new automated ultrasound method for placental volume measurement that allows the placenta to be imaged from multiple angels, using 2D ultrasound with position tracking and artificial intelligence in form of a convolutional neural network (CNN) for automatic image segmentation. Methods: Tracked 2D ultrasound images of the placenta were acquired from 43 pregnant women at gestational week 27. We manually segmented the placenta in 2583 images, and used the images from 38 placentas to train a CNN for automatic segmentation. The CNN was then applied to the images from the remaining 15 placentas, and the results were combined with tracking data to generate a 3D model and calculate the placental volume. The volumes of these 15 placentas were also estimated from manually segmented 2D and 3D ultrasound images. The methods were compared to MRI. Results: The CNN achieved an F1-score of 0.84. The correlation with MRI-based placental volume measurements was similar for automatically segmented (ICC 0.64, 95% CI 0.21–0.86), manually segmented (ICC 0.66, 95% CI 0.24–0.87) and 3D (ICC 0.63, 95% CI 0.20–0.86) ultrasound. Ultrasound measured smaller volumes than MRI, but this systematic error was less pronounced for the methods based on 2D ultrasound, probably because more placental tissue was captured in the images. Conclusions: Placental volume can be automatically measured using tracked 2D ultrasound and a trained CNN. The method may be further improved. In 2023 we have also been working with the following manuscripts that are now published : Placental size at gestational week 27 and 37: The associations with pulsatility index in the uterine and the fetal-placental arteries. Kulseng CPS, Sommerfelt S, Flo K, Gjesdal KI, Peterson HF, Hillestad V, Eskild A. Placenta. 2024 Jan;145:45-50. doi: 10.1016/j.placenta.2023.11.014. Epub 2023 Dec 4. Placental size at gestational week 36: Comparisons between ongoing pregnancies and deliveries. Peterson HF, Eskild A, Sommerfelt S, Hillestad V. Acta Obstet Gynecol Scand. 2024 Jan;103(1):85-92. doi: 10.1111/aogs.14700. Epub 2023 Oct 30. Automatic placental and fetal volume estimation by a convolutional neural network. Kulseng CPS, Hillestad V, Eskild A, Gjesdal KI. Placenta. 2023 Mar 24;134:23-29. doi: 10.1016/j.placenta.2023.02.009. Epub 2023 Feb 27.

Two-dimensional (2D) placental ultrasound measurements – the correlation with placental volume measured by MRISimple 2D ultrasound measurements of the placenta were poorly correlated with placental volume and cannot be used as surrogate measures of placental volume, this is probably because of a large variation between pregnancies in intrauterine placental shape.1) Two-dimensional (2D) placental ultrasound measurements – the correlation with placental volume measured by MRI (to be submitted for publication) Objectives: Information about placental size in ongoing pregnancies may aid the identification of pregnancies with increased risk of adverse outcome. Placental volume can be measured using magnetic resonance imaging (MRI). However, such method is usually not easily available in antenatal care. Ultrasound is the diagnostic tool of choice in pregnancy. Therefore, we studied whether simple 2D ultrasound placental measurements were correlated with placental volume measured by MRI. Methods: We examined 104 unselected ongoing pregnancies in gestational week 27, using both ultrasound and MRI. The ultrasound measurements included placental length, width and thickness. Placental volume was measured using MRI. The correlation between each 2D placental ultrasound measurement and placental volume was estimated by applying the Pearson’s correlation coefficient (r). Results: Mean placental length was 17.2 cm (SD 2.1 cm), mean width was 14.7 cm (SD 2.1 cm), and mean thickness was 3.2 cm (SD 0.6 cm). Mean placental volume was 536 cm3 (SD 137 cm3). The 2D ultrasound measurements showed poor correlation with placental volume (placental length; r = 0.27, width; r = 0.37, and depth r = 0.13). Conclusions: Simple 2D ultrasound measurements of the placenta were poorly correlated with placental volume and cannot be used as surrogate measures of placental volume, this is probably because of a large variation between pregnancies in intrauterine placental shape. 2) Automatic placental and fetal volume estimation by a convolutional neural network (conditionally accepted paper) Introduction: We aimed to develop an artificial intelligence (AI) deep learning-based algorithm to efficiently estimate placental and fetal volumes based on magnetic resonance (MR) scans. Methods: Manually annotated images from an MRI sequence was used as input to the neural network DenseVNet. We included data from 193 normal pregnancies at gestational week 27 and 37. The data were split into 173 scans for training, 10 scans for validation and 10 scans for testing. The neural network segmentations were compared to the manual annotation (ground truth) using the Dice Score Coefficient (DSC). Results: The mean ground truth placental volume at gestational week 27 and 37 was 571 cm3 (Standard Deviation (SD) 293 cm3) and 853 cm3 (SD 186 cm3), respectively. Mean fetal volume was 979 cm3 (SD 117 cm3) and 2,715 cm3 (SD 360 cm3). The best fitting neural network model was attained at 30,000 training iterations. The neural network estimated mean placental volumes at gestational week 27 to 615 cm3 (SD 155 cm3) (DSC 0.894 (SD 0.039), and to 840 cm3 (SD 264 cm3) at gestational week 37 (DSC 0.900 (SD 0.023). Mean fetal volumes were 949 cm3 (SD 140 cm3) and 2,619 cm3 (SD 269 cm3), with mean DSC of 0.950 (SD 0.009) and 0.970 (SD 0.003). The time spent for volume estimation was reduced from 60-90 minutes by manual annotation, to less than 10 seconds by the neural network. Conclusion: The correctness of the neural network volume estimation is comparable to human performance; the efficiency is substantially improved. Other presentations: Placental volume measurements using tracked 2D ultrasound. Torgrim Lie, Karianne Sagberg, Helene Fjeldvik Peterson, Vigdis Hillestad, Anne Eskild og Lars Eirik Bø

Placental volume in gestational week 27 measured by three-dimensional ultrasound and magnetic resonance imagingWe found poor to moderate correlation between 3D ultrasound and MRI placental volume measurements. Generally, 3D ultrasound measured smaller placental volumes than MRI, suggesting that 3D ultrasound failed to visualize the entire placenta. Our findings may hopefully contribute to the improvement of ultrasound methods for placental measurements.Abstract Introduction: Ultrasound is the diagnostic tool of choice in pregnancy. We lack valid ultrasound methods for placental size measurements. Our aim was therefore to compare three-dimensional (3D) ultrasound with magnetic resonance imaging (MRI) for measurements of placental volume. Material and methods: We measured placental volume by 3D ultrasound and MRI in 100 unselected pregnancies at 27 weeks of gestation (25+4 -28+4 weeks). The 3D ultrasound acquisitions were analyzed offline, and the placental outline was manually traced using the virtual organ computer-aided analysis (VOCAL) 30° rotational technique. The MRI examinations included a T2-weighted gradient echo sequence in the sagittal plane, with 5-mm slices through the entire uterus. The placental outline was manually traced in each slice. The correlation between 3D ultrasound and MRI placental volumes was estimated by intraclass correlation coefficients. Bland-Altman analysis was applied to visualize systematic bias and limits of agreement, in which the ratio MRI placental volume/3D ultrasound placental volume was plotted against the average of the two methods. Results: The intraclass correlation coefficient between 3D ultrasound and MRI measurements was 0.49 (95% confidence interval 0.33-0.63). In general, 3D ultrasound measured smaller placental volumes (median 373 cm3 , interquartile range 309-434 cm3 ) than MRI (median 507 cm3 , interquartile range 429-595 cm3 ) and the systematic bias was 1.44. The 95% limits of agreement between the two methods were wide (0.68-2.21). Presentasjoner 2021: Sagberg K, Eskild A, Sommerfelt S, Gjesdal KI, Higgins L, Borthne A, Hillestad V. Placental volume in gestational week 27 measured by three dimensional (3D) ultrasound and magnetic resonance imaging (MRI). European Congress of Obstetrics and gynecology. Athen. Podium. Peterson H, Eskild A, Sommerfelt S, Gjesdal KI, Borthne A, Mørkrid L, Hillestad V. Intrauterin placental volume and placental volume relative to fetal volume: percentiles based on magnetic resonance imaging. European Congress of Obstetrics and gynecology. Athen. Podium. Sagberg K, Lie T, Peterson H, Bø LE, Gjedal KI, Hillestad V, Eskild A. Måling av morkakevolum ved bruk av frihånds ultralydopptak og maskinlæring. Norsk gynekologisk forenings årsmøte. Tønsberg. Poster. Peterson H, Eskild A, Sommerfelt S, Gjesdal KI, Borthne A, Mørkrid L, Hillestad V. Intrauterin placental volume and placental volume relative to fetal volume: percentiles based on magnetic resonance imaging. European Congress of Obstetrics and gynecology. Norsk gynekologisk forenings årsmøte. Tønsberg. Podium. Utmerkelser 2021: "Best presentation award" EBCOG 2021. Sagberg K, Eskild A, Sommerfelt S, Gjesdal KI, Higgins L, Borthne A, Hillestad V. Placental volume in gestational week 27 measured by three dimensional (3D) ultrasound and magnetic resonance imaging (MRI). Invent2 prisen. The Baby Segmentations Tool. Manuskripter under utarbeiding: Sagberg K, Eskild A, Sommerfelt S, Halle TK, Higgins L, Hillestad V. Simple biometric placental measurements by ultrasound - any relation with placental volume? In progress. Peterson H, Eskild A, Sommerfelt S, Gjesdal KI, Borthne A, Mørkrid L, Hillestad V. Percentiles of intrauterine placental volume and placental volume relative to fetal volume: A prospective magnetic resonance imaging study. In review.

Placental volume in gestational week 27 measured by three dimensional (3D) ultrasound and magnetic resonance imaging (MRI)We found poor to moderate correlation between 3D ultrasound and MRI placental volume measurements. Generally, 3D ultrasound measured smaller placental volumes than MRI, suggesting that 3D ultrasound failed to visualize the entire placenta. Our findings may hopefully contribute to the improvement of ultrasound methods for placental measurements.Introduction: Ultrasound is the diagnostic tool of choice in pregnancy. We lack valid ultrasound methods for placental size measurements. Our aim was therefore to compare three-dimensional (3D) ultrasound with magnetic resonance imaging (MRI) for measurements of placental volume. Methods: We measured placental volume by 3D ultrasound and MRI in 100 unselected. pregnancies at 27 weeks of gestation (25+4 - 28+4 weeks). The 3D ultrasound acquisitions were analysed offline, and the placental outline was manually traced using the virtual organ computer-aided analysis (VOCAL) 30° rotational technique. The MRI examinations included a T2-weighted gradient echo sequence in the sagittal plane, with 5 millimetre (mm) slices through the entire uterus. The placental outline was manually traced in each slice. The correlation between 3D ultrasound and MRI placental volumes was estimated by Intraclass Correlation Coefficients. Bland Altman analysis was applied to visualise systematic bias and limits of agreement, in which the ratio (MRI placental volume/3D ultrasound placental volume) was plotted against the average of the two methods. Results: The Intraclass Correlation Coefficient between 3D ultrasound and MRI measurements was 0.49 (95% confidence interval 0.33-0.63). In general, 3D ultrasoundmeasured smaller placental volumes (median 373 cm3 , interquartile range 309-434 cm3) than MRI (median 507 cm3 , interquartile range 429-595 cm3) and the systematic bias was 1.44. The 95% limits of agreement between the two methods were wide (0.68-2.21). New projects: The results presented above have initiated a collaborative effort with SINTEF to further develop ultrasound as a placental diagnostic tool. Funded by the Norwegian Research Council. The PLAVO study has also recently received funding from the OlsoMet. A PhD candidate who will develop methods for estimations of placental perfusion. Publications: Placental volume in gestational week 27 measured by three dimensional (3D) ultrasound and magnetic resonance imaging (MRI). Karianne SAGBERG, Anne ESKILD, Silje SOMMERFELT, Kjell Inge GJESDAL, , Lucy E. HIGGINS, Arne BORTHNE, and Vigdis HILLESTAD. Acta Obstetricia et Gynecologica Scandinavica in press. In preparation Intrauterine placental and fetal volume during the last part of pregnancy and placental to fetal ratio: Percentiles based on magnetic resonance imaging. Helene. F. PETERSON, Anne ESKILD., Silje SOMMERFELT, Kjell-Inge GJESDAL, Arne S. BORTHNE , Lars MØRKRID MD, and Vigdis HILLESTAD.

Kan ultralyd brukes til å måle placentavolum i pågående svangerskap? En valideringsstudie av ultralyd mot MRMorkakevolument målt med tre dimensjonal (3D) ultralyd er korrelert med MR volumen. Men vi tror ultralyd-metoden må forbedres før den tas i klinisk bruk.Bakgrunn: Avvikende placentavekt har blitt assosiert med uheldige svangerskapsutfall. Å kunne måle placentas størrelse i pågående svangerskap kan derfor potensielt diagnostisere risikosvangerskap. Flere ultralydmetoder har blitt brukt til å måle placentas volum i pågående svangerskap. Imidlertid har ingen studert hvor valide disse metodene er. Mål: Å validere to- og tredimensjonal (2D, 3D) ultralyd som metode for å måle placentavolum. Vi brukte placentavolum målt med MR som gullstandard. Metode: Vi undersøkte placentaer hos 104 tilfeldig utvalgte gravide med 2D og 3D ultralyd og MR i svangerskapsuke 27 (variasjonsbredde 25+4- 28+4). For 2D-ultralyd målte vi lengde, bredde og tykkelse av placenta og regnet ut volumet basert på formelen for en ellipse. For 3D ultralyd brukte vi Virtual Organ Computer-aided AnaLysis (VOCAL) 30° -teknikk, og bildeopptakene ble analysert offline for beregning av volum. Ved MR ble det gjort bildeopptak med 5 millimeter tykke snitt gjennom hele livmoren. Placentas omriss ble tegnet manuelt i hvert snitt og volumet ble beregnet basert på snitt-tykkelsen. Vi sammenliknet gjennonsnittsvolum og studerte også korrelasjon ved de forskjellige metodene ved å bruke intraclass correlation coffisient (ICC) og Bland Altman plot med limits of Agreement (LoA). Resultater: Gjennomsnittlig placentavolum var mindre målt med 2D ultralyd (423 kubikkcentimeter (cm3) variasjonsbredde 167-735cm3) og 3D ultralyd (378cm3, variasjonsbredde 223-597cm3) sammenliknet med MR (536cm3, variasjonsbredde 323-999cm3).2D og 3D ultralydmål av placentavolum viste relativ lav korrelerlasjon med placentavolum målt med MR, ICC var henholdsvis 0.38 (95 CI 0.21-0.54, p<0.000) og 0.46 (95 CI 0.29-0.60, p<0.0001). Konklusjon: 2D og 3D ultralyd er foreløpig ikke gode nok metoder for å måle placentavolum i svangerskapsuke 27. Presentasjoner 2019 Eskild A. Placental volume measured by ultrasound and magnetic resonance imaging (MRI) in pregnancy week 27, a method validation study. Society of Epidemiological Reseach. Minneapolis. Podium. Peterson H, Eskild A, Sagberg K, Sommerfelt S, Gjesdal KI, Borthne A, Hillestad V. Hvordan vokser morkaken i svangerskapet? En MR (magnetisk resonans) studie. Poster, NGF, Bodø 23-25 okt 2019. Sagberg K, Sommerfelt S, Gjesdal KI, Higgins L, Borthne A, Eskild A, Hillestad V. Kan ultralyd brukes til å måle placentavolum i pågående svangerskap? En valideringsstudie av ultralyd ved bruk av magnetisk resonans (MR) som gullstandard. Podium, NGF, Bodø 23-25 okt 2019. Sommerfelt S, Flo K, Hillestad V, Sagberg K, Eskild A. Er pulsatilitetsindeks (PI) i uterinarteriene knyttet til størrelsen på placenta relativt til fosteret? Poster, NGF, Bodø 23-25 okt 2019. Manuskripter under utarbeiding Sommerfelt S, Flo K, Gjesdal KI, Borthne A, Hillestad V, Sagberg K, Eskild A. The association between Doppler pulsatility indices (PI) and the placental to fetal size ratio – an ultrasound and MRI study. In progress. Sagberg K, Sommerfelt S, Gjesdal KI, Higgins L, Borthne A, Eskild A, Hillestad V. Placental volume measured by ultrasound and magnetic resonance imaging (MRI) in pregnancy week 27, a method validation study. In progress Peterson HF, Eskild, A, Sagberg K, Sommerfelt S, Gjesdal KI, Borthne A, Hillestad V. Growth rate of the placenta and the placental size relative to the fetal size in ongoing pregnancies, an MRI study. In progress

Kan 3D ultralyd bukes til å måle placentavolum? En studie av intra-rater reliabilitet.Nye metoder i forskning og i klinikk forutsetter at lik undersøkelse av samme pasient gir samme resultat (høy reliabilitet). Vi studerte intra-rater reliabilitet for placentavolum ved bruk av 3D ultralyd, og vi fant høy grad av reliabilitet .Presentasjon på årsmøtet for Norsk gynekologisk årmsøte, Oslo, oktober 2018: Kan 3D ultralyd bukes til å måle placentavolum? En studie av intra-rater reliabilitet. Sommerfelt S¹², Halle T², Eskild A¹², Siafarikas F¹² ¹Kvinneklinikken, Akershus universitetssykehus. ²Institutt for klinisk medisin, Det medisinske fakultet, Universitetet i Oslo. Introduksjon. Kunnskap om placentavolum i svangerskapet kan være viktig for å identifisere fostre som har et dårlig intrauterint miljø. Stor eller liten placenta relativt til fosterets størrelse kan tyde på dårlig miljø. Tredimensjonal (3D) ultralyd er en ny metode for å måle placentavolum. Nye metoder i forskning og i klinikk forutsetter at lik undersøkelse av samme pasient gir samme resultat (høy reliabilitet). Vi studerte intra-rater reliabilitet for placentavolum ved bruk av 3D ultralyd. Metode. Vi inkluderte 28 gravide kvinner med singelton svangerskap i 2. trimester. Gjennomsnittlig svangerskapslengde var 19+1 uker. Kvinnene ble undersøkt to ganger av samme undersøker med en GE Voluson ultralydmaskin. Ved hver undersøklese ble det tatt et 3D-bilde av placenta. Etterpå ble hvert placentabilde manuelt avmerket (tracet) to ganger ved bruk av GE 4D-view software, noe som ga beregnet placentavolum. Intra-rater reliabilitet ble estimert ved bruk av intraclass correlation coefficient (ICC) og Limits of agreement (LoA). Resultater. Gjennomsnittlig placentavolum ved første tracing av begge placentabildene var henholdsvis 231,33 (SD 50,73) cm³ og 226,30 (SD 71,66) cm³. ICC var 0,89 med nedre LoA på -70,98 og øvre LoA på 81,04. Gjennomsnittsvolumet av begge tracingene var på 231,49 (SD 50,93) cm³ for det første placentabildet, og 229,30 (SD 64,81) cm³ for det andre. ICC var 0,92 med nedre LoA på -61,23 og øvre LoA på 65,61. Konklusjon. Måling av placentavolum ved bruk av 3D ultralyd gir utmerket intra-rater reliabilitet. LoA var vide da placenta ble tracet første gang. LoA ble redusert ved bruk av gjennomsnittet av to tracinger per bilde. All datainnsamling i studien er utført. Inklusjon og oppfølging gjennom svangerskapet av 250 gravide kvinner. Manus er under utarbeiding: K. Sagberg, S. Sommerfeldt, A. Eskild, and V. Hillestad V. Placental volume measured by ultrasound and MRI in pregnancy week 27, a method comparison study.

Measurement of placental volume using 2D and 3D ultrasound as compared to MRI – a pilot studyWe have performed a pilot study, comparing placental volume as estimated by using 2D, 3D ultrasound and MRI. The results were presented at an international conference on placental research in Manchester, spring 2017.Measurement of placental volume using 2D and 3D ultrasound as compared to MRI – a pilot study lSiljeSommerfelt1VigdisHillestad3LucyHiggins4KarianneSagberg1TuvaHalle12Kjell-IngeGjesdal32Arne SigurdBorthne32AnneEskild1,2 1 Division of Obstetrics and Gynecology, Akershus University Hospital, Lørenskog, Norway 2 Institute of Clinical Medicine, Oslo University, Oslo, Norway 3 Department of Diagnostic Imaging, Akershus University Hospital, Lørenskog, Norway 4 Maternal and Fetal Health Research Centre, Institute of Human Development, University of Manchester, Manchester, UK Objectives: In this pilot study, we assessed the feasibility of validating placental volume measurement by using two-dimensional (2D) and three-dimensional (3D) ultrasound, with magnetic resonance imaging (MRI) as the gold standard. Methods: Five women with a healthy singleton pregnancy in week 26-27 were included. MRI was performed on a 1.5T system using balanced fast field echo (bFFE) sequences through the uterus. MRI placental volumes were measured based on sagittal slices. 2D and 3D ultrasound examinations were performed by four different examiners. For estimating placental volume with 2D ultrasound, we measured placental length, width and depth, and we assumed an elliptical placenta for calculation. Placental volume was calculated with 3D ultrasound Virtual Organ Computer-aided AnaLysis (VOCAL). Six 30° rotational planes of the placenta were traced manually by all examiners by using Voluson 4D-View software. We present the means and the ranges of the measurements. Results: Placental volumes obtained by MRI were generally larger than volumes obtained by 3D ultrasound. Placental volume measurement with 2D and 3D ultrasound had wide interobserver ranges (Table 1). When different observers traced the same 3D image, the ranges were narrower (Table 2). Table 1: Placental volume (cm³). Mean, minimum and maximum values, and ranges as estimated by 2D and 3D ultrasound and MRI. Woman MRI 2D Ultrasound 3D Ultrasound Mean Range Mean Range 1 626 566 (471-661) 190 397 (318-484) 166 2 n/a 413 (356-470) 114 336 (253-468) 215 3 516 594 (352-765) 413 380 (200-514) 313 4 440 403 (354-494) 140 479 (367-636) 269 5 499 325 (233-448) 215 265 (211-315) 105 Table 2: 3D ultrasound images obtained by four different examiners. Mean estimated placental volume and ranges (cm³) as traced offline manually. Each of the four images was traced by 3-4 observers, a total of 13 traces per placenta. Examiner ID Woman 1 2 3 4 Mean Range Mean Range Mean Range Mean Range 1 370 (347-403) 55 326 (318-338) 20 463 (453-484) 31 419 (381-467) 86 2 338 (271-386) 115 271 (258-282) 24 294 (253-361) 108 414 (339-468) 129 3 363 (312-393) 81 496 (462-514) 51 206 (200-214) 14 436 (408-475) 67 4 428 (367-470) 103 425 (404-453) 48 568 (526-636) 110 492 (398-605) 207 5 285 (261-315) 54 252 (230-254) 40 221 (211-227) 16 295 (274-305) 31 Conclusion: Further validation and intra- and interobserver reliability studies of ultrasound measurement of placental volume should be performed before ultrasound measures of placental volume are used in research. Such studies should be designed to minimize missed placental tissue when imaging the placenta. DOI: http://dx.doi.org/10.1016/j.placenta.2017.07.217