Using Active Learning Techniques for Analysis of Photomicroscopy Images.

Processing biomedical image data, specifically microscopic data from a blood sample, where it is necessary to identify and classify leukocytes, is a complex issue. Manual processing of such data is time-consuming and can be subjective and prone to human error. Therefore, this work aims to analyze current processing techniques and available tools to simplify and automate this process. The paper includes an overview of the biomedical data used, and describes the analysis, the selection of the relevant data, a proposal for a solution using the general methods and overall automation of the process, an implementation of the solution, and an evaluation of its benefits in further research. Logistic regression was used to predict the leukocyte type, in combination with the active learning approach. The work results in a solution that simplifies and mainly speeds up the processing of microscopic data for leukocyte identification and classification by removing the human error and subjectivity. The proposed solution achieved classification accuracy of 92% in both training and testing dataset (Raabin-WBC) and 84% on vakidation dataset (BCCD), without any changes in the model parameters. This solution may find application in diagnosing various diseases such as blood cancers, infections, autoimmune diseases, or allergies.