Color-based Fuzzy Classifier for automated detection of cancerous cells in histopathological analysis

Abstract

In usual clinical practice, grading of cancer tissues is often done by an expert histopathologist based on their analysis of micro-level architectural features of the cancerous tissue specimen as well as level of presence of certain protein molecules in the specimen. This process of assessment of the level of presence of protein molecules is extremely subjective due to human dependence and limitations and, therefore, causes large inter-expert and sometimes even intra-expert variability and potentially adding noise to the process of selecting the treatment regime that the patient is put on. Quantification of immunohistochemical (IHC) stains used to highlight, in general, healthy and malignant tissues, is critical for an objective assessment of any type of cancer's histopathological specimens. In addition, pathologist's workload is also a serious problem in any hospital, where disease load and mortality statistics are increasing. The impact of not having an objective method for diagnosis and the lack of pathologists will lead to delay in diagnosis and misdiagnosis that can harmfully affect patient treatment and survival. In this paper, an algorithmic approach is presented in order to quantify and then classify based on the same type of information that a human expert would utilize, i.e. color. The subjectivity of color detection is embedded in the design as Fuzzy Classifier based on a range of colors rather than specific color values. The procedure can be automated to classify, and then further quantify for any specific measure, such as number of specific type of cells present, etc... This is expected to reduce the load on the pathologists and increase the quality of diagnosis.