Kebelakangan ini, semakin ramai penyelidik menjalankan penyelidikan tentang aplikasi sistem pintar dalam bidang perubatan. Tujuan projek ini ialah membina sebuah sistem pintar yang berupaya mengelaskan tahap barah pangkal rahim kepada tiga tahap, iaitu, normal, Low grade Squamous Intra-epithelial Lesion (LSIL), atau High grade Squamous Intra-epithelial Lesion (HSIL) berdasarkan ciri-ciri pada imej ThinPrep. Keupayaan dan kesesuaian rangkaian neural untuk digunakan dalam sistem pengelasan tahap barah pangkal rahim akan dikaji. Dua jenis rangkaian neural iaitu rangkaian Perceptron Berbilang Lapisan (MLP) dan rangkaian Perceptron Berbilang Lapisan Hibrid (HMLP) akan dibina. Prestasi kedua-dua rangkaian neural ini akan dibandingkan untuk menentukan rangkaian neural yang paling sesuai untuk sistem pengelasan ini. Rangkaian MLP dibina dengan MATLAB akan dilatih dengan tiga algoritma yang berlainan, iaitu, Algoritma Perambatan Balik, Levenberg-Marquardt dan aturan Bayesian. Manakala HMLP akan dibina dengan Borland C++ Builder dan dilatih dengan Algoritma Ralat Ramalan Rekursif Ubahsuai. Pretasi rangkaian neural dibuat berdasarkan ketepatan, sensitiviti, spesifisiti, nilai ramalan positif, dan nilai ramalan negatif. Rangkaian HMLP menunjukkan prestasi yang lebih baik berbanding dengan rangkaian MLP. Rangkaian HMLP berjaya mencapai 96.68% ketepatan, 97.16% sensitiviti, 95.65% spesifisiti, 99.42% nilai ramalan positif, dan 81.48% nilai ramalan negatif. Hasil akhir ialah sebuah perisian yang berupaya mengelaskan tahap barah pangkal rahim dengan tepat, jitu, cepat, dan mudah.
__________________________________________________________________________________________
Throughout the years, many researchers have been conducted on the potential applications of Artificial Intelligence (AI) in medical field. The purpose of this project is to develop an intelligent system to classify cervical pre-cancerous stage into normal, Low grade Squamous Intra-epithelial Lesion (LSIL), or High grade Squamous Intra-epithelial Lesion (HSIL) based on the features obtained from ThinPrep Images. The capability and suitability of neural networks as intelligent classification will be investigated. In this project, conventional Multilayered Perceptron (MLP) network and Hybrid Multilayered Perceptron (HMLP) network will be developed and their performances are compared to yield the most suitable network that will be used to model the classification system. The MLP will be developed using MATLAB and trained with Back Propagation, Levenberg Marquardt, and Bayesian Regularization learning algorithms. While the HMLP with Modified Recursive Prediction Error learning algorithm will be developed using Borland C++ Builder. The performance of neural networks was done based on accuracy, sensitivity, specificity, positive predictive value, and negative predictive value. The final result shows the HMLP network performed better classification when compare with the MLP. HMLP was able to achieve 96.98% of accuracy, 97.16% of sensitivity, 95.65% of specificity, 99.42% of positive predictive value, and 81.48% of negative predictive value. The final product of this project is a software system that is capable to classify cervical pre-cancerous stage with high accuracy, high applicability, fast and cheap.