Ultrasound-based radiomics technology in fetal lung texture analysis to predict neonatal respiratory morbidity | Scientific Reports – Nature.com | Candle Made Easy

patients

Between July 2018 and October 2020, 2047 routine ultrasound images of the fetal lungs (either right or left lung) were obtained from 2047 women with singleton pregnancies of gestational age (GA) 27 years+3 until 42+0 weeks. All participating women who took part in the study gave their written informed consent to the use of ultrasound images and clinical data. All methods explained herein were performed in accordance with relevant guidelines and regulations and approved by the Ethics Committee of Fudan University Obstetrics and Gynecology Hospital (2018-73) together with the study protocol. Of these, 731 babies with GA 28+3–37+6 Weeks were delivered within 72 h of hospital ultrasound examination. Using the same inclusion criteria from previous studies, the final cohort included 295 women with singleton pregnancies with a total of 295 fetal lung ultrasound images. The flow chart for the study population is shown in Fig. 3. Gestational age was determined from last menstrual period and verified by first-trimester dating ultrasound (crown-rump-length).

figure 3
figure 3

Flow chart of study population selection. NRM neonatal respiratory morbidity.

Pregnancy complications included GDM and PE. GDM was diagnosed with a 75 g oral glucose tolerance test at weeks 24-28. diagnosed week of pregnancy27. Preeclampsia and gestational hypertension are characterized by the recurrence of hypertension (> 140 mmHg systolic or > 90 mmHg diastolic) after 20 weeks of gestation28.

Analysis of neonatal clinical data was supervised by a neonatal physician. NRM included respiratory distress syndrome (RDS) or transient tachypnea of ​​the newborn (TTN). The diagnosis of IBS and TTN is based on symptoms, signs and a radiological examination7.29. Diagnostic criteria of IBS: tachypnea, snoring, chest wall retraction, nasal dilation, the need for supplemental oxygen, and the appearance of chest radiographs led to admission to the NICU for respiratory support. Diagnostic criteria for TTN: mild or moderate respiratory distress (isolated tachypnea, infrequent snoring, mild retraction) and a chest x-ray (if performed) showing alveolar and/or pulmonary interstitial effusion and abnormal pulmonary vascular patterns.

Ultrasound imaging and segmentation

All ultrasound images were obtained during routine prenatal ultrasound examinations within 72 hours prior to delivery. Below, the training set images were acquired by Radiologist 1, who has more than 10 years of experience in obstetric and gynecological ultrasound imaging, using a WS80A ultrasound system (Samsung, Korea). The frequency of the CA1-7A probe was 1-7 MHz, with a center frequency of 4.0 MHz. The test set images were acquired by Radiologist 2 with 3 years of experience in obstetric and gynecological ultrasound imaging using a VOLUSON E8 Ultrasound System (GE, USA). The C1-5-D probe frequency was 2-5 MHz, with a center frequency of 3.5 MHz.

A detailed description of the standard image acquisition protocol and the method of manual (freehand) delineation is fully described in a previous study25: In short, the standard images of the fetal lungs require: On an axial section of the fetal thorax at the level of the four-chamber cardiac view, the settings (depth, gain, frequency and harmonics) were adjusted to ensure that at least one of the lungs had no obvious acoustic Shadowing by the fetal ribs. All images were inspected for image quality control and saved in DICOM format (.dcm) for offline analysis. Manual (freehand) delineation was performed in each fetal lung by two radiologists (radiologists A and B), and square delineation (40 × 40 pixels) was performed by radiologist B, choosing one side of the fetal lung sites with great care Ensure that only the lung tissue has been delineated, avoiding blood vessels, costal shadows and the pulmonary capsule as shown in Figure 4. Radiologist A’s segmentation results were used to generate the model, while Radiologist B’s segmentation and quadratic delineation results were used to check the stability of the model.

figure 4
figure 4

Ultrasound images of fetal human lungs with regions of interest defined. (a,a1,a2,a3) are images of the training set. (b,b1,b2,b3) Are pictures of the test set. (a1,b1) Manual delineation (radiologist A) of each lung. (a2,b2) Manual delineation (radiologist B) of each lung. (a3,b3) Square representation (40 × 40 pixels) of each lung. (a,a1,a2,a3) image of the left lung at 36+1 weeks in women with preeclampsia (PE). The cesarean section was performed 2 days after the ultrasound, and the baby was diagnosed with transient neonatal tachypnea. The risk probability derived from the model is 0.829 (> 0.5). (b,b1,b2,b3) Image of the left lung at 34+0 weeks in women with gestational diabetes (GDM). The cesarean section was performed immediately after the ultrasound, and the baby was diagnosed with respiratory distress syndrome. The risk probability derived from the model is 0.843 (> 0.5).

Evaluation of radiomics and machine learning

The research process is shown in Fig. 5.

Figure 5
Figure 5

Workflow of the fetal lung texture analysis system based on ultrasound-based radiomics technology. Stage I: The US image of the fetal lung (four-chamber view) was manually segmented. Phase II: 430 high-throughput radiomic features were extracted from each segmented image. Then features were selected by permuting out-of-bag data features of a random regression forest. And the prediction model was built with RUSBoost (Random Undersampling with AdaBoost). Finally, the risk probability of NRM in each fetal lung image was obtained and classified into the high-risk group or the low-risk group. Level III: According to the results of the confusion matrix, the performance of the prediction model was evaluated using sensitivity (SENS), specificity (SPEC), accuracy (ACC), and area under the receiver-operating-characteristics (ROC) curve. ROI area of ​​interest, US Ultrasonic, NRM neonatal respiratory morbidity, sense Sensitivity, spec specificity, acc Accuracy, ROC Operating characteristics of the receiver.

All feature extractions and image classifications were performed using Matlab R2018a and Toolbox Classification (Mathworks, Inc, Natick, Massachusetts, USA).

Univariate analysis was used to describe the differences in characteristics between different categories. The t-test was performed on 430 continuous radiomics features each25, including 15 morphological, 73 texture, and 342 wavelet features. The χ2 The test was performed for two categorical clinical features, gestational age and pregnancy complications. P Score < 0.05 indicated a significant difference.

The feature extraction method for analyzing each ROI has already been reported25. First, the importance of high-throughput radiomics features per fetal lung image was assigned to selected features by permuting out-of-bag data features of the random regression forest. If a feature is influential, permuting its values ​​would affect model error checking with out-of-bag data. The more important a feature is, the greater its impact30. As a result, 20 radiomic features (2 textural features and 18 wavelet features) and 2 clinical features (GA and pregnancy complications) were selected for classification, presented in Table 4. The stability of selected radiomic features depending on different delineations (manual delineation by radiologists A and B and quadratic delineation) was analyzed with ICC (2, 1)31. Then, diagnostic performance in predicting neonatal respiratory morbidity was compared depending on different traits, including clinical traits (GA and pregnancy complications), radiomics traits, and the combination of clinical and radiomics traits. For the clinical features, a Support Vector Machine (SVM) classifier was used for classification. By adjusting the cost of misclassification in different categories, the classifier can focus on the positive samples. For radiomics functions and the combination of clinical and radiomics functions with the high sample imbalance and small sample size RUSBoost (random undersampling with AdaBoost)32 was used to build the model. Finally, the risk probability of NRM in each fetal lung image was obtained, which was the predicted score normalized to the range of 0-1 by the softmax function of the RUSBoost. The limit of the model was 0.5. Fetal lungs with a risk probability greater than 0.5 were classified as high-risk and less than 0.5 as low-risk. All classifier parameters were tuned using 10-fold bootstrap cross-validation, and the decision tree was used as a base learner for RUSBoost.

Table 4 List of high-throughput sonographic features.

The model’s predictive performance was evaluated in terms of sensitivity (SENS), specificity (SPEC), accuracy, PPV, NPV, and AUC.

Leave a Comment