A machine learning algorithm was able to distinguish patients with neuromyelitis optica spectrum disorder (NMOSD) from those with relapsing-remitting multiple sclerosis (RRMS) or peripheral neuropathy and healthy controls.

The study, published in the International Journal of Molecular Science, showed that the algorithm was able to produce sensitivity and specificity of 100% in the test set when discriminating NMOSD, RRMS, peripheral neuropathy, and the control group using Fourier transform infrared spectroscopy (FTIR) of patients’ blood sera.

“The coupling of infrared spectroscopy of sera to machine learning is a promising cost-effective, rapid and reliable differential diagnosis tool capable of helping to gain valuable time in patients’ treatment,” the authors said.


Continue Reading

Despite success differentiating between diseases, no clear signal was able to discriminate between the 3 different serostatuses of NMOSD—those with antibodies to aquaporin-4, antibodies to myelin oligodendrocyte glycoprotein (MOG), and those with neither type of antibodies. All 3 serostatuses were able to be individually discriminated from the patients with RRMS, however.

Read more about NMOSD differential diagnosis.

The study authors said they are currently in the process of creating an application that could automate the steps of data acquisition, post-processing, and machine learning classification to allow medical staff to quickly perform analysis without the need for specific expertise. The algorithm would also provide a “spectrum score” and “classification score” that would provide details on the quality of the recorded spectra as well as the disease diagnosis.

“FTIR spectroscopy coupled to a random forest classifier can offer a fast (data acquisition takes less than 30 min) and cost-effective additional tool to improve the diagnosis and differentiation of RRMS and NMOSD, regardless of serostatus,” the authors said.

A total of 235 sera were collected for the study from patients including 60 with NMOSD, 60 with RRMS, 35 with peripheral neuropathy, and 80 healthy controls. Of those samples, 208 were used for training and validation of the algorithm while the remaining 27 (6 NMOSD, 6 RRMS, 10 control, and 5 peripheral neuropathy) were used as the final validation set.

The FTIR signals from each of the sera were collected in the 4000 to 700 cm-1 spectral range. The signals from several scans were averaged for each patient before undergoing baseline correction, normalization, and generation of the second derivative. These processed signals were then used as inputs for a random forest algorithm which utilized 100 random decision trees to classify the data.

Reference

El Khoury Y, Gebelin M, de Sèze J, et al. Rapid discrimination of neuromyelitis optica spectrum disorder and multiple sclerosis using machine learning on infrared spectra of sera. Int J Mol Sci. 2022;23(5):2791. doi:10.3390/ijms23052791