Efficient image texture analysis and classification for prostate ultrasound diagnosis

Abstract

An efficient, integrated image textural analysis and classification of transrectal prostate ultrasound images into clusters potentially representing cancerous or normal tissue areas is presented. Preliminary image texture analysis has shown the potential for doubled diagnosis accuracy from 38-42% for prostate cancer with current clinical methods, to 88-92%. In addition, image texture analysis makes prostate cancer locating possible for more precise, less invasive biopsy/treatment, instead of 6-way random biopsy. However, the initial image texture analysis on a mini VAX could take 8 days CPU time per image, i.e., more than 5 months for 20 cross-sections per patient. Over the last 10 years, we have improved the processing from 8 days to less than 10 seconds per image on a PC. The approach is based on Haralick's textural features and the Minimum Squared Error (MSE) clustering algorithm. The Java Textural Analysis/Classification (JTAC) application developed as part of this project offers significant reduction in run time, potentially allowing more accurate, objective diagnoses to be performed within clinical settings, and allows the investigation of parameters associated with textural and clustering processes. Using this integrated approach, specific results for several cases are tested and general conclusions are drawn.