Effective improvement of cancer diagnostics and prognostics by computer-assisted cell image analysis

摘要

DNA Image Cytometry is a method for the early diagnosis and prognosis of cancer.It exploits, as a biomarker for cancer, the DNA content of morphologically suspiciousnuclei measured from digital images. Therefore, the identification of these suspiciousnuclei in a microscopic inspection is a crucial step of the method. Until now, this task had to be performed by a pathological expert who required, onthe average, 40 minutes per slide - prohibitive for a wide-spread routine application. This thesis presents image processing algorithms for accomplishing this task automatically,the core component being classifiers which are capable of distinguishingmorphologically abnormal nuclei from normal nuclei, other types of nuclei, and artifacts. These algorithms were integrated into a software package, and a workflowwhich loads the tedious work onto the machine leaving only critical tasks to theresponsible expert. This provides an overall solution, which was evaluated in threeclinically relevant applications: the identification of cancer cells in nuclei from serouseffusions and from brush biopsies of the oral cavity, and grading the malignancy ofprostate cancer biopsies. The developed solution reduces the workload for the expert to 5 minutes per slide.As compared the previous visual selection of nuclei, in addition both the diagnosticaccuracy and prognostic validity are increased.