Continuous class pattern recognition for pathology, with applications to non-hodgkin's follicular lymphomas

摘要

Continuous class pattern recognition is a new analytic method that differs from existing techniques by recognizing and exploiting the continuous relationship among classes along diagnostic scales. Continuous class pattern recognition was applied to a wide variety of image processing features extracted from lymph node biopsy images that were digitized using a Coulter diff3/50 automated research microscope. The resultant classifiers correctly subtyped 89% of a set of 37 follicular lymphomas, compared to individual pathologist rates that ranged from 57% to 81%. This study demonstrates that continuous class pattern recognition can significantly reduce this diagnostic error rate.