Mid-infrared spectroscopy as a real-time diagnostic tool for chronic endometritis: A preliminary study

Research question: Can attenuated total reflection-Fourier transform infrared spectroscopy combined with machine learning techniques be used to develop a real-time diagnostic modality for chronic endometritis by analysing endometrial biopsies obtained during hysteroscopy?

Design: Women undergoing hysteroscopy for infertility assessment were enrolled in this prospective study from January 2020 to March 2021. Endometrial biopsies were evaluated using a spectrophotometer, and subsequently via histopathology, including immunohistochemical staining for the multiple myeloma oncogene-1 (MUM-1). Spectroscopy analyses of the positive and the negative chronic endometritis groups were compared across various cut-offs of MUM-1 positive cells per 10 high-power fields (HPF). Machine learning techniques were used to build discrimination models with Matlab and Unscrambler software packages.

Results: Fifty-four women were recruited. Four of the 54 measured spectra were discarded due to high measurement noise. Fifty biopsies were finally assessed, and a discriminant model was developed using the Principal Component Analysis and Linear Discriminant Analysis techniques (machine learning). The model was evaluated for accuracy using different MUM-1 cut-offs. Distinct spectral disparities (p < 0.05) were observed between specimens negative or positive for chronic endometritis. When employing a cut-off of > 7 MUM-1 cells/10 HPF, the model differentiated positive and negative chronic endometritis with 84 % accuracy, 88.8 % sensitivity and 82.9 % specificity.

Conclusions: Our findings in this preliminary study support the potential availability of a bedside diagnostic tool based on a model developed using spectroscopy coupled with machine learning. This allowed us to diagnose chronic endometritis in real-time with 84% accuracy, facilitating immediate initiation of appropriate treatment.