Robust Band Profile Extraction Using Constrained Nonparametric Machine-Learning Technique

Shadab Khan, João M. Sanches, Rodrigo Ventura

Abstract: A typical characteristic of images of bone marrow cells taken during mitosis is poor quality. This renders the task of extraction of accurate band profile, representative of intensity distribution over each chromosome, more challenging. A robust method is hence required to tackle this problem. An algorithm was thus developed, which estimates a single-line medial axis, the basis for computation of band profile. Medial axis was generated by computing a final prediction, using primary and secondary predictions obtained by a nonparametric machine learning algorithm trained with data from chromosome's skeleton, and geometrical properties of medial axis, respectively. Experiments were performed using the LK1 dataset. The algorithm was found capable of estimating a satisfactory single-line medial axis. Band profile obtained was found to be a good representation of intensity levels in different regions of chromosomes. Additionally, this algorithm is robust in terms of growing a very small seed region into desired medial axis and also handling highly irregular chromosomes.