Efficient Leukocyte Classification Based On Multimodal Light-source Images Using Ensemble-free Network

Litao Yang, Deval Mehta, Zongyuan Ge

Keywords: WBCs, fluorescence, bright-field, efficient, multimodal, ensemble

Abstract: Classifying white blood cells (leukocytes) in a blood sample is essential for diagnosing the health condition of a person. Conventionally, this is accomplished in a central clinical laboratory with trained experts and sophisticated blood cell counter systems. Recently, there has been an increase in developing machine learning and deep learning techniques based on blood smear and fluorescent images for this task. In this work, we present an approach based on multimodal fluorescence and bright-field images of blood samples which are exposed to excitation wavelengths of different light sources. To this end, we collect a multimodal (four modalities) dataset of 6,700 white blood cells present in peripheral blood. Despite the multimodal nature of our dataset, we propose a low complexity ensemble-free deep learning network for performing leukocyte classification. In our proposed approach, multiple separated subnetworks of a single network can learn features from modality specific images. This enables our approach to provide an almost on par classification performance while having 4x fewer parameters than that of a traditional ensemble system employed for the same task. Our proposed ensemble-free architecture can achieve an overall accuracy of 96.15% for 5-part differential leukocyte classification while having only 1.3M parameters. We believe that our proposed approach can also help with developing an efficient point-of-care (POC) solution for leukocyte classification especially for resource poor environments.

Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.

Paper Type: both

Primary Subject Area: Detection and Diagnosis

Secondary Subject Area: Application: Other

Confidentiality And Author Instructions: I read the call for papers and author instructions. I acknowledge that exceeding the page limit and/or altering the latex template can result in desk rejection.