Leukemia recognition with evolutionary vision and knowledge transfer

Abstract

Leukemia is a health problem that affects the world population causing thousands of kills. Visual analysis is a diagnostic method required for leukemia detection. Evolutionary vision models are useful to understand the solutions found because they make it possible to identify how the image recognition process develops. In this work, we analyze the performance of an evolutionary vision model named brain programming, in which we use a multi-class classifier embedded into the model, to lead the evolutionary process. Furthermore, we use a type of knowledge transfer to improve the recognition task. Results suggest that the knowledge transfer allow an adequate model performance to solve the problem of leukemia recognition. In addition, the structure of the model provides some degree of explainability to the approach, and the structure of the solutions found is amenable to interpretation, which is a desirable aspect of the leukemia recognition problem.