Digital Pathology: How Far Are We from Automated Tissue-Based Diagnosis?

Abstract

Background Tissue based diagnosis (TBD) includes all diagnostic pro-cedures that are performed on human tissue for disease classification and treatment. Its computerized information analysis is called Digital Pathology. Herein we will discuss the present stage of IT tools and the assumed clinical perspectives on medical performance and treatment. Theory Basically, TBD investigates the function and structures of bio-logical meaningful individual units, such as macromolecules, genes, nuclei, cells, vessels, and organs. All functions are bound to structures that ensure reliable and effective infor-mation and energy exchange. Disturbance of structures induces less effective or complete loss of functions. The complex interactions at molecular biological level (macro-molecules) and their continuous reproduction require exten-sive computations in addition to the sophisticated biochem-ical analysis systems. Nearly all assessable information is of visual nature or can be visualized. Thus, image content analysis applied in a sophisticated manner might be one key procedure to assist human image interpretation or to even replace it. Image content information includes information that can be derived from predefined functional units (objects), their spatial arrangement (structure), pixel derived features prior of after image transformations (texture), and syntac-tic compositions of objects or of pixel based primitives (syntactic structure analysis). Statistically significant clusters can represent either new biological significant units (e.g., tubular arrangement of specific (endothelial) cells forming a vessel, spatial composition of cells of different nature (cellular sociology) forming a bronchus with assumed participation of endogenous lectins [1]) or other new items such as entropy flow charts and diffusion densities. All these parameters form a powerful set of image information features. They can be considered to be independent from each other and calculated independently for their specific clinical significance (disease association). Present Status The development of whole slide image scanners, their imple-mentation into laboratory and hospital information systems, and development of internet based open access image mea-surement systems permit the construction of automated disease classifiers with inbuilt image quality control and monitoring. The prerequisites include image standardization (of gray value range, distribution, magnification in relation to object measurements, etc.), detection of regions of inter-est (ROI), standardized image transformation procedures,