Inferensi Hasil Karakterisasi FTIR Kadar Glukosa Darah Manusia Menggunakan Artificial Neural Network (ANN)

Abstract

Pengukuran kadar glukosa darah menggunakan metode standar pengukuran kadar glukosa darah konvensional yang melibatkan reagen cenderung lebih mahal dan memiliki banyak kekurangan, oleh karena itu dibutuhkan metode baru yang memotong reagen dan dapat lebih efektif daripada metode konvensional. Penelitian ini bertujuan untuk membangun, mengembangkan, dan menguji performa metode pengukuran kadar glukosa darah non-invasive tanpa melibatkan reagen dengan menggunakan metode kecerdasan buatan atau Artificial Neural Network (ANN) serta memberikan informasi mengenai kandidat lampu yang layak dalam pengukuran kadar glukosa darah melalui metode akurasi yang meliputi analisis Root Mean Square Error (RMSE) serta menggunakan metode uji diagnostik yang meliputi analisis Error Grid Analysis (EGA), analisis sensitivitas, analisis spesifisitas, analisis akurasi diagnosis, dan analisis Number Needed to Diagnose (NND). Analisis korelasi absorbansi terhadap akurasi dan uji diagnosispun dilakukan. Kandidat lampu yang memiliki akurasi berupa RMSE dan EGA yang baik yaitu pada rentang panjang gelombang 1041 nm sampai 1063 nm dan 1063 nm sampai 1086 nm dengan RMSE berturut-turut 9,6 mg/dl dengan absorbansi 0,68 dan standar deviasi 0,05 dan 11,5 mg/dl dengan absorbansi 0,66 dan standar deviasi 0,04. Clarke EGA dan Parkes EGA pada dua kandidat tersebut berada dalam zona A dan zona B. Kandidat lampu berdasarkan analisis uji diagnosis yang telah dilakukan menghasilkan 1 kandidat lampu terbaik yaitu kandidat lampu yang berada pada rentang panjang gelombang 1351 nm sampai 1428 nm dengan sensitivitas bernilai 87% yang meliputi absorbansi sebesar 0,66 dan standar deviasi absorbansi sebesar 0,04, spesifisitas bernilai 80% yang meliputi absorbansi sebesar 0,73 dan standar deviasi absorbansi sebesar 0,053, akurasi diagnosis bernilai 83% yang meliputi absorbansi sebesar 0,64 dan standar deviasi absorbansi sebesar 0,04, dan Needed to Diagnose (NND) bernilai 1,5 repetitions yang meliputi absorbansi sebesar 0,64 dan standar deviasi absorbansi sebesar 0,04. Hal tersebut menunjukkan bahwa metode kecerdasan buatan berupa Artificial Neural Network (ANN) menghasilkan kandidat panjang gelombang yang layak untuk direalisasikan dalam pengukuran kadar glukosa darah manusia baik ditinjau dari sudut pandang akurasi pengukuran maupun kemampuan dalam melakukan diagnosis status glikemia.

Measurement of blood glucose levels using the standard method of measuring conventional blood glucose levels involving reagents tends to be more expensive and has many drawbacks, therefore a new method is needed that bypasses the reagent and can be more effective than conventional methods. This study aims to build, develop, and test the performance of a non-invasive method of measuring blood glucose levels without involving reagents using the Artificial Neural Network (ANN) method as well as providing information about suitable lamp candidates in measuring blood glucose levels through the accuracy method which includes Root Mean Square Error (RMSE) as well as using diagnostic test methods which include and Error Grid Analysis (EGA) analysis, sensitivity analysis, specificity analysis, diagnosis accuracy analysis, and Number Needed to Diagnose (NND) analysis. Absorbance correlation analysis on accuracy and diagnostic tests were also carried out as an effort to test the performance of the ANN method. Lamp candidates that have good RMSE and EGA accuracy are in the wavelength range of 1041 nm to 1063 nm and 1063 nm to 1086 nm with RMSE 9,6 mg/dl and 11,5 mg/dl, respectively. Clarke EGA and Parkes EGA in the two candidates are in zone A and zone B. In each of these wavelength ranges the absorbance values ​​are 0,68 and 0,66 with standard deviation ranges of 0,05 and 0,04. Lamp candidates based on diagnostic test analysis resulted in the 1 best lamp candidates. Lamp candidates based on the analysis of diagnostic tests that have been carried out resulted in 1 best lamp candidate, namely lamp candidates in the wavelength range of 1351 nm to 1428 nm with 87% sensitivity which includes 0,66 absorbance and 0,04 standard deviation of absorbance, 80% specificity covering 0,73 absorbance and 0,053 standard deviation of absorbance, the accuracy of diagnosis 83% which includes 0,64 absorbance and 0,04 standard deviation of absorbance, and 1,5 repetitions Number Needed to Diagnosis (NND) which includes 0,64 absorbance and 0,04 absorbance standard deviation. This shows that the artificial intelligence method in the form of an Artificial Neural Network (ANN) produces a suitable wavelength candidate for measuring human blood glucose levels both in terms of accuracy and ability in diagnosing glycemic status.