A Preliminary Fuzzy Model to Identify Abnormal Cervical Smears Using Fourier Transform Infrared Spectroscopy

Abstract

Cervical cancer accounts for 6% of female malignancies with an estimated 4,000 new cases and 1,200 deaths per year in England and Wales. Current methods of screening are difficult to perform, error prone and subjective. A technique known as Fourier transform infrared spectroscopy (FT-IR) can be used to observe differences in vibrational modes of molecules in tissue proteins, and has been shown to hold promise for the early diagnosis of malignant changes. Accurate prediction of potential malignancy is thought to involve analysis of key parameters such as peak ratios and/or peak location at certain known important frequencies. A small database of cases with confirmed clinical diagnosis was used to investigate an initial fuzzy model of expertise capable of assessing the likely presence of malignancy. Two established, conventional rule-induction techniques (C4.5 and CN2) were used to induce a provisional set of rules which were then translated into fuzzy rule equivalents. The principles of FT-IR and its application to cervical cancer diagnosis, the development process of the initial fuzzy model of expertise, and provisional evaluation results are presented. Early results indicate that this model will need to be refined through knowledge elicitation with clinical experts and by fuzzy model tuning on more cases with known diagnosis.