Introduction: A five part differential leucocyte count is provided by a hematology analyser with abnormalities being detected as flags which are then later on reviewed and confirmed by a pathology trainee or pathologist based on the difficulty level of that individual case or knowledge of the particular trainee. Peripheral smear examination is considered the gold standard and is time consuming and subjective. 

Methods: Using a self prepared 1500 leishman stained leucocyte image dataset the machine learning programme tensor flow from google using image intensity, histogram and convolutional neural network was trained and this was put to test on 80 random leucocyte images.

Results: Only 65% concordance was obtained on 5 tier leucocyte differential count. However 95% concordance was achieved by using a two tier leucocyte differential classification of polynuclear and mononuclear cells. 

Conclusion: Larger dataset of images are needed before image analysis using this model can be used routinely to substitute or as addon to routine peripheral smear examination.

Keywords: Image Classifier; Differential Leucocyte Count; TensorFlow; Peripheral  Smear.