Future directions in quantitative pathology: digital knowledge in diagnostic pathology.

The ancient Pythagorean philosophers believed in numbers. Numbers were the key to an understanding of the world and—because they were eminently practical people—to prosperity. As we enter the information age we hold similarly strong beliefs in the benefits of digital technology: numerical representation oVers objectivity, discreteness, permanence, analytic evaluation. There has always been the explicit hope that quantification of knowledge in histopathology will eventually lead to better medical practice. It is not an unfounded hope. The introduction of quantitative micromorphometric procedures to diagnostic histopathology has shown the potential of information technology based methods. There is the value of objective mensuration with the prospect of standardisation of diagnostic assessment. There is the ability to establish the significance of small and early alterations. There is the proven ability to extract diagnostic information that is not reliably recognised by visual microscopic assessment. There is the ability to measure rapidly, in numerical terms, the eYcacy of chemopreventive intervention and thus the ability to test a large number of potentially eVective compounds. Over all, there has been outstanding progress. Yet it may be time to take stock and to ask where we stand and where future eVorts might be productive. In fact, we must ask whether we are using to a maximum the information that digital imagery oVers. Questions arise as to whether the methodology underlying the quantitative analysis provides the information that we need, and even whether it is appropriate for some of the problems encountered in diagnostic and prognostic histopathology. Image processing, diagnostic information extraction, and quantitative assessment have for years relied on methods borrowed from the material sciences and from statistical analysis, often following the lead of research in “pattern recognition.” In fact, mathematical/statistical procedures form the bulk of analytic tools in quantitative histopathology and they are indeed appropriate and immensely useful in many instances. However, one must not forget that statistical procedures are designed to assess groups of observations and not to characterise individuals. What the clinician wants to know from a pathologist, though, is not only the specific diagnostic assessment but what the prognosis and best management plan is for a particular patient. There is an inherent mismatch between the design objective of the analytic procedures and the desired utility. As one examines our analytic arsenal of mathematical/statistical procedures and the needs of diagnostic and prognostic histopathology it becomes evident that additional analytic approaches will have to be developed. Are we asking all the questions that we should?