Automated Detection of Blood Leukemia Using Image Pre-Processing and Machine Learning

The present work proposes a technique for blood cancer (leukemia) classification based on machine learning. Manual classification of blood cancer is very challenging and prone to errors and therefore machine learning approaches are employed to perform the task. Out of machine learning algorithms, deep learning has emerged as the suitable candidate for analyzing such large and complex datasets with enormous divergences. The proposed work presents a feature selection and training algorithm based on the RESNET-50 Deep Neural Network for image classification. The RESNET employs the skip connection among the hidden layers so as to reduce chances of overfitting and vanishing gradient. The performance of the proposed system has been evaluated in terms of the error and accuracy. It is found that the proposed technique achieves almost 97.9% classification accuracy which is better than previous work for same dataset (91.84%).