Data processing in histopathology.

Abstract

A histopathologist performs in diagnosis at least five fundamental tasks: he abstracts and mentally integrates features of tissue structure; he relates these patterns to information drawn from memory and other records; he makes inferences from all these data; he then forms and finally communicates a description. In any consideration of the ways in which computer technology can affect this sequence, we are immediately faced with the fact that the primary process, the recognition of pattern, is an exceedingly complex and largely unanalysed cerebral activity. Automatic recognition of pattern is still in its infancy, and although a beginning has been made in the limited and relatively simple fields of cytology and chromosome analysis towards the removal of the human brain from the scene, it is proposed to confine the present discussion to the remaining aspects of the process, namely, the formation, storage, recall, and communication of descriptions. The process of describing and making inferences from the patterns which we recognize, although open to logical analysis, is again a complex business. In practice, the simpler and more obvious the pattern, the fewer the words we employ. In the simplest cases of all we may proceed directly to the inference and apply a classifying label without more ado. On the other hand, where the pattern is not clearly seen, our descriptions are more consciously analytical and form the basis for an assessment of probabilities. This part of the process is comparable to clinical diagnosis and in principle is open to the techniques of propositional calculus (Feinstein, 1967), numerical taxonomy, and multivariate analysis (Baron and Fraser, 1965; Hayhoe, Quaglino, and Doll, 1964) which are being used experimentally in that field. However, the more immediate benefits to histopathology which are likely to be obtained from the application of modern data processing methods lie in the direction of the storage and retrieval of data on a large scale. Before discussing methods it seems pertinent to consider the reasons for wishing to embark on any system of data collection. I would suggest that these are both local and general in nature. At the local level, the linking of the histopathology reports with the remainder of the patients' records is the primary need. The problem at this point becomes part of the much larger and more difficult problem of the organization of the medical record for data processing. The potential benefits to pathology of the capacity to make correlations with other laboratory and with clinical data need no emphasis, although we are as yet far from a successful solution. Secondly there is the indexing of the local collection of reports and sections: this is the traditional field for most of the data processing that has been practised in histopathology to the present time. To turn to the motives which could lead to the collection of data on a larger scale, these include the formation of large data banks for reference and research, and the collection of data for statistical analysis of morbidity and mortality. Examples of this kind of use already exist in the shape of the various regional cancer collections, and a strong case can be made for the extension of the principle to other types of pathological data. Furthermore, I would suggest that the term 'pathological data' could in this context well be extended to include other classes of related information whose significance is enhanced when it is reviewed in the mass. Anthropometric data are of this kind: for example, it is possible to calculate total body fat and muscle and skeletal mass from simple postmortem measurements (Alexander, 1964). Again, the details of traumatic death could provide a valuable source of facts in the field of accident prevention. The problem is largely one of organization; at its heart lies the need for a free exchange of information between centre and periphery. As we know all too well, there is a natural and proper resistance to the labour of collecting and supplying information without any visible return. Before any pathologist could take part in a larger scheme with enthusiasm, he would reasonably expect free and rapid access to centrally stored information, this to include slides as well as documents. Given the necessary central organ-