Therapeutic monitoring of digoxin with surface-enhaced raman spectroscopy

Digoxin is a heart glicoside which is used for treatment of heart failure and atrial fibrillation. It is mostly used in clinics but it needs to be monitorised about dosage and toxicity effects because it has a very narrow therapeutic window (1,0-2,5 nmol/L). Surface-enhanced Raman spectroscopy which gives more sensitive results than Raman spectroscopy is used to detect low dosage drugs and the sample which is difficult to analyze. After analysis, some important information is obtained about samples by using chemometric approach which is the discipline for investigation of chemical data statistically and mathematically. In that thesis, a well-known drug which has a narrow therapeutic window and needs to be monitorized during therapy, digoxin, is detected quantitavively and classified qualitatively with SERS and chemometric analyses in human serum. Spectral data were analyzed by using 5 chemometric methods and approaches, NB, NN, RF, SGD and SVM which are well-known machine learning methods in artificial intelligence. Raw data must be pre-processed to reduce noise, if any, to make peaks more distinct before applying the method, or to achieve higher accuracy rates.Although the classification algorithms were implemented without pre-processing, they were also pre-processed with Gaussian smoothing, Savitzky-Golay filter, baseline correction, normalization (vector, minimum-maximum and area normalization), PCA, spectrum variation and asymmetric least squares smoothing approaches. In the first experiment group, 5 chemometric approaches were used to distinguish mixture with digoxin (D) and mixture without digoxin (S). Between that 5 approaches, neural network (NN) gave the best classification and detection result (Accuracy=0,971 AUC=0,954). In the second experiment group, again the same 5 chemometric approaches were used to distinguish mixture containing high and toxic dosage digoxin (Y) and mixture containing therapeutic dosage digoxin (T). Between that 5 approaches, neural network (NN) and SGD gave the best classification and detection results (NN with CA=0,936 AUC=0,964 and SGD with CA=0,936 and AUC=0,930). So, a new, fast, effective and cheap alternative method was developed to help both distinguish patients who take or not take digoxin during therapy and detect and decide about digoxin level is toxic or therapeutic in clinics.