Facilitating Leucocyte Count Using Deep Learning A Paradigm Shift

Aim:

          The aim of this paper is to develop an automatic white blood cell classification system using deep learning techniques. Specifically, the objective is to create a model capable of accurately classifying white blood cells (leucocytes) to aid in the diagnosis of various hematologic diseases, including anaemia and leukaemia. Unlike previous approaches that relied on transfer learning with models trained on the ImageNet dataset, this study proposes a novel deep learning model that is built from scratch, tailored to handle the unique characteristics of the white blood cell dataset used in this application.

Abstract:

       The classification of white blood cells or leucocytes has become highly crucial for the diagnosis of anaemia, leukaemia and many other hematologic diseases. The density of WBCs in our blood stream provides a glimpse into the state of our immune system and any potential risks we might be facing. The main objective of this paper is to develop an automatic white blood cell classification system using deep learning. Most of the models proposed for this application so far has used transfer learning by fine tuning the ‘State of the art” models like ResNet50. But all these models were trained on ImageNet dataset which is completely different from the dataset used in this application. So in this study, we have proposed a deep learning model for the white blood cells classification.

Proposed System:

          In this paper, we propose a novel white blood cell classification system using the ResNet50 architecture, which is a deep Convolutional Neural Network known for its state-of-the-art performance in various computer vision tasks. However, instead of employing transfer learning from pre-trained models on unrelated datasets like ImageNet, we fine-tune the ResNet50 specifically on our white blood cell dataset. This approach aims to leverage the advantages of the ResNet50 architecture while addressing the limitations observed in the existing system.

Advantage:

         Fine-tuning ResNet50 on the relatively small white blood cell dataset can help mitigate over fitting issues compared to training from scratch. The pre-trained weights of ResNet50 serve as a useful starting point, capturing general features and allowing the model to focus on learning domain-specific features without losing generalization capability.