Classifying cells from light microscopic bit features by binary logic. Application to grade neuronal injury in...

Degradative cellular processes within neurons were analyzed and graded by a conceptually new approach that decoupled the process of subcellular feature analysis and grading, the latter being based on the severity of observed deterioration in subcellular features. Rather than evaluate the cell's histologic image in a panoramic manner, investigators were required to give simple yes-no decisions about the presence or absence of a specific pathologic feature (bit feature as opposed to panoramic analysis). Multiple elementary decisions create a binary representation of the cell image that can be easily handled and analyzed using computer techniques to generate all possible unique phenotypes of the scanned neuronal population. In the first application of this method, however, the number of possible phenotypes were reduced by imposed a priori logic on the separation scheme focusing on a single cellular structure (the nucleus) that was followed through the stages of structural decay. We experimentally validated four neuronal types of five theoretical possibilities when three nuclear bit features were used in typing. Grading of neuronal injury for groups of normal, ischemic, and ischemic and reperfused rats into two, three, and four categories are reported. The consistency at which the method can be implemented was assessed by calculating the mean and standard deviation of the reconciled typing decisions given by the four investigators. The group of the four investigators showed less than 2, 3, and 5% error when grading cells from control, ischemic, and ischemic-reperfused animals, respectively.