Abstract:Recurrent spontaneous miscarriage refers to the repeated loss of two or more clinically detected pregnancies occurring within 24 weeks of gestation. No identifiable cause has been identified for nearly 50% of these cases. This group is referred to as idiopathic recurrent spontaneous miscarriage (IRSM) or miscarriage of unknown origin. Due to lack of robust scientific evidence, guidelines on the diagnosis and management of IRSM are not well defined and often contradictory. This motivates us to explore the vibrational fingerprints of endometrial tissue in these women. Endometrial tissues were collected from women undergoing IRSM ( n = 20) and controls ( n = 20) corresponding to the window of implantation. Attenuated total reflectance-Fourier transform infrared (ATR-FTIR) spectra were obtained within the range of 400–4000 cm −1 using Agilent Cary 630 FTIR spectrometer. Raman spectra were also generated within the spectral window of 400–4000 cm −1 using Thermo Fisher Scientific, DXR Raman spectrophotometer. Based on the limited molecular information provided by a single spectroscopic tool, fusion strategy combining Raman and ATR-FTIR spectroscopic data of IRSM is proposed. The significant features were extracted applying principal component analysis (PCA) and wavelet threshold denoising (WTD) and fused spectral data used as input into support vector machine (SVM), adaptive boosting (AdaBoost) and decision tree (DT) models. Altered molecular vibrations associated with proteins, glutamate, and lipid metabolism were observed in IRSM using Raman spectroscopy. FTIR analysis indicated changes in the molecular vibrations of lipids and proteins, collagen dysregulation and impaired glucose metabolism. Combination of both spectroscopic data using mid-level fusion (MLF: 92% using AdaBoost and DT models) and high-level fusion (HLF: 92% using SVM models) methods showed improved IRSM classification accuracy as compared to individual spectral models. Our results indicate that spectral fusion technology hold promise in enhancing diagnostic accuracy of IRSM in clinical settings. Validation of these findings in a larger patient population is underway. Machine learning algorithms applied to spectroscopic fusion data for prediction of idiopathic recurrent spontaneous miscarriage. (A) Recruitment of two groups of subjects – (i) idiopathic recurrent spontaneous miscarriage (IRSM) cases and (ii) controls. (B) Raman and Fourier transform infrared (FTIR) spectroscopic data were generated from endometrial tissue of IRSM and controls collected during implantation window. Data were subjected to chemometric analysis, such as unsupervised hierarchical cluster analysis (HCA) and supervised orthogonal partial least squares-discriminant analysis (OPLS-DA) which showed optimized discrimination between the two groups. (C) Raman and FTIR data were subjected to mid-level and high-level fusion where data matrix were extracted using principal component analysis (PCA) and wavelet denoising. (D) The selected features were used as input into adaptive boosting, decision tree and support vector machine models and the best performing models selected.

Classification of idiopathic recurrent spontaneous miscarriage using FTIR and Raman spectroscopic fusion technology