Ultrasound based radiomics model for assessment of placental function in pregnancies with preeclampsia

Abstract

The goal of our research is to elucidate and better assess placental function in rats with preeclampsia through an innovative application of ultrasound-based radiomics. Using a rat model induced with L-NAME, we carefully investigated placental dysfunction via microstructural analysis and immunoprotein level assessment. Employing the Boruta feature selection method on ultrasound images facilitated the identification of crucial features, consequently enabling the development of a robust model for classifying placental dysfunction. Our study included 12 pregnant rats, and thorough placental evaluations were conducted on 160 fetal rats. Distinct alterations in placental microstructure and angiogenic factor expression were evident in rats with preeclampsia. Leveraging high-throughput mining of quantitative image features, we extracted 558 radiomic features, which were subsequently used to construct an impressive evaluation model with an area under the receiver operating curve (AUC) of 0.95. This model also exhibited a remarkable sensitivity, specificity, accuracy, positive predictive value, and negative predictive value of 88.7%, 91.5%, 90.2%, 90.4%, and 90.0%, respectively. Our findings highlight the ability of ultrasound-based radiomics to detect abnormal placental features, demonstrating its potential for evaluating both normative and impaired placental function with high precision and reliability.

Keywords: Placenta; Preeclampsia; Radiomics; Rats; Ultrasound.

Fig. 1

The procedures of utrasound-based radiomics.

Fig. 2

Comparison of fetal rat and placental development. (a) Overall embryonic development of rats in the control group; (b) Overall embryonic development of rats in the L-NAME group, which showed a higher mortality rate among fetal rats and even absorption of ‘fetus’; (c) The fetal rat of control groups. (d) The fetal rat of L-NAME group. (e) The comparison of fetal weight between the control group and the L-NAME group; (f) The comparison of placental weight between the control group and the L-NAME group; (g) The comparison of fetal placental weight ratio between the control group and the L-NAME group, The fetal placental weight ratio is indicative of placental functionality.

Fig. 3

Pathological morphology between control group and L-NAME group. (A) (C) The microscopic structure of the control group; (B) (D) The microscopic structure of the L-NAME group; The control and the L-NAME group showed a significant structural difference in the rat labyrinth zone. (A) (B) Magnification: 400×; (C) (D) Magnification: 200×.

Fig. 4

Protein expression levels in placental tissue between control group and L-NAME group. (A) Immunoblot analysis of the control and L-NAME groups of the strip charts; (B) Placental relative protein expression levels of sFIt-1 in the control and L-NAME groups; (C) Placental relative protein expression levels of PlGF in the control and L-NAME groups; (D) Placental relative protein expression levels of sFlt-1/PlGF in the control and L-NAME groups. *P < 0.05.

Fig. 5

(a), (b), (c) Visualization of the distribution of ultrasound image features in rat placenta samples; (d), (e) Single-factor analysis of the radiomics features of the normal and dysfunction groups; (f) Visualization of features contributing to decision making in the rat placental function assessment model. SHAP: Shapley Additive explanations, for each sample model produces a predictive value, Shap value is the value assigned to each feature in the sample; FSN: Feature serial number.

Fig. 6

Confusion matrix and ROC curve plots of rat placental function assessment model. (a) Confusion matrix; (b) Receiver operator characteristic curve.