Using blood data for the diferential diagnosis and prognosis of motor neuron diseases: a new dataset for machine learning applications

Early diferential diagnosis of several motor neuron diseases (MNDs) is extremely challenging due to the high number of overlapped symptoms. The routine clinical practice is based on clinical history and examination, usually accompanied by electrophysiological tests. However, although previous studies have demonstrated the involvement of altered metabolic pathways, biomarker-based monitoring tools are still far from being applied. In this study, we aim at characterizing and discriminating patients with involvement of both upper and lower motor neurons (i.e., amyotrophic lateral sclerosis (ALS) patients) from those with selective involvement of the lower motor neuron (LMND), by using blood data exclusively. To this end, in the last ten years, we built a database including 692 blood data and related clinical observations from 55 ALS and LMND patients. Each blood sample was described by 108 analytes. Starting from this outstanding number of features, we performed a characterization of the two groups of patients through statistical and classifcation analyses of blood data. Specifcally, we implemented a support vector machine with recursive feature elimination (SVM-RFE) to automatically diagnose each patient into the ALS or LMND groups and to recognize whether they had a fast or slow disease progression. The classifcation strategy through the RFE algorithm also allowed us to reveal the most informative subset of blood analytes including novel potential biomarkers of MNDs. Our results show that we successfully devised subject-independent classifers for the diferential diagnosis and prognosis of ALS and LMND with remarkable average accuracy (up to 94%), using blood data exclusively.