Fingernail Diagnostics: Advancing type II diabetes detection using machine learning algorithms and laser spectroscopy

• Published in: Microchemical journal Vol. 201; p. 110762
• Main Authors: Rehan, Imran, Rehan, Kamran, Sultana, Sabiha, Ur Rehman, Mujeeb
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.06.2024
• Subjects: Ensemble learning; Laser-induced breakdown spectroscopy; Machine learning; Stack learning; type-II diabetes; Stack learning; Ensemble learning; type-II diabetes; Laser-induced breakdown spectroscopy; Machine learning
• ISSN: 0026-265X
• DOI: 10.1016/j.microc.2024.110762

[Display omitted] •Non-invasive controlled diabetes mellitus diagnostics using laser-induced breakdown spectroscopy.•Employed Q-switched Nd: YAG laser for spectral analysis of fingernail samples.•Elemental variations in diabetic versus non-diabetic fingernails.•Machine learning enhances spectral differentiation for diabetes diagnosis.•Stack learning model proves effective for precise diabetic sample classification. Prolonged Type-II diabetes disrupts the typical function of the heart, kidneys, nerves, blood vessels, bones, and joints. Type-II diabetes gradually modifies the inherent material properties and structural integrity of tissues, while extended periods of hyperglycemia result in the chronic deterioration of tissue quality. With this intention, machine learning and Artificial Intelligence (AI) have been employed in recent years. In the current work, nanosecond pulsed laser-induced breakdown spectroscopy (LIBS) has been utilized to explore the impact of type II-controlled diabetes mellitus upon the chemical contents of fingernails. Discrimination was executed on 80 nail clippings, with 40 from individuals with diabetes and 40 from control subjects. From these fingernail samples, a total of 4800 LIBS emission spectra were acquired. The differentiation between individuals with and without diabetes was initially accomplished through the utilization of principal component analysis (PCA) on LIBS spectral data and subsequently integrated into a novel machine-learning model. The proposed categorization framework for a non-invasive scheme utilized seven distinct classifiers and employed cross-validation procedures to assess and compare the outcomes. The classification outcomes were encouraging, achieving satisfactory accuracy, precision, sensitivity, specificity levels, and F1-score of 96 %, 99.9 %, 96.7 %, 99.9 % and 96.8 respectively. The preliminary finding demonstrates that utilizing LIBS spectra of fingernails in conjunction with machine learning can be a valid method for classifying individuals as either diabetic or nondiabetic, making it a feasible approach for screening purposes.