Image statistics and data mining of anal intraepithelial neoplasia

• Published in: Pattern recognition letters Vol. 29; no. 16; pp. 2189 - 2196
• Main Authors: Ahammer, H., Kröpfl, J.M., Hackl, Ch, Sedivy, R.
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.12.2008; Elsevier
• Subjects: Applied sciences; Classification; Data mining; Exact sciences and technology; HIV; Image statistics; Information theory; Information, signal and communications theory; Neoplasia; Signal and communications theory; Signal representation. Spectral analysis; Signal, noise; Telecommunications and information theory; HIV; Data mining; Neoplasia; Image statistics; Classification; Second order; Automatic classification; High precision; Data processing; Malignant tumor; Signal classification; Learning; First order; Cancer
• ISSN: 0167-8655; 1872-7344
• DOI: 10.1016/j.patrec.2008.08.008

Anal intraepithelial neoplasia (AIN) is a precancerous condition of growing concern, due to the strong interrelation of AIN with infections caused by human papillomaviruses (HPV) and HIV. Several HPV-subtypes induce a variety of tumorous skin lesions and cause different stages of dysplasia and even cancer. The histological classification of AIN is becoming more and more important in clinical practice, due to increasing HPV infection rates throughout human population. Histological slices of anal tissues are commonly classified by individual inspections with all the unavoidable differences of the training status and variances of the individual. Therefore, a quantitative classification method including the calculations of first order as well as second order image statistical parameters in combination with data mining was developed. The results of several classifiers were compared to each other and it turned out that at least two classifiers had very high correct classification rates with very low errors. So it was possible to classify the distinct grades of AIN with high accuracy. The quantitative approach has the potential to minimize individual classification errors significantly and it will enable the establishing of a quantitative screening technique.