Combining elemental analysis of toenails and machine learning techniques as a non-invasive diagnostic tool for the robust classification of type-2 diabetes

Described for the first time is the use of elemental analysis of diabetic toenails and machine learning techniques for the robust classification of type-2 diabetes. Aluminum, Cs, Ni, V and Zn concentrations in toenails were found to be significantly (p < 0.05) different between healthy volunteers and type-2 diabetes patients. Seven different machine learning algorithms were then studied to develop a non-invasive diagnostic method using concentrations of twenty-two elements in toenails, and personal information such as age, gender and smoking history as features. Models were enhanced through feature selection and two different ensembling strategies. The performance of forty-six distinct machine learning models were compared on resampled training data and testing data. A random forest model, trained with concentrations of Al, Ba, Ca, Cr, Cs, Cu, Fe, Mg, Mn, Ni, P, Pb, Rb, S, Sb, Se, Sn, Sr, V and Zn (µg g−1), as well as information on age, gender and smoking history, had an area under the receiver operating characteristic curve (AUC) of 0.73 on the training data, and correctly predicted seven out of nine test samples (including control and disease), with an AUC of 0.90. The results at this stage of the research prove the concept of combining elemental analysis of toenails and machine learning techniques for non-invasively diagnosing type-2 diabetes. With proper sample collection and shipping, mobility-limited patients may be able to mail toenail samples for analysis and monitor their type-2 diabetes over time. A health clinic equipped with common instrumentation, software and trained algorithms similar to those used in the present study may be able to serve a large number of patients from across the world.