Deep Learning for Malaria Screening
Brown Bag Lecture by Dr. Sivarama Krishnan Rajaraman | 3/20/2018 11:00AM – 12PM | 7th Floor Conference Room, Bldg 38A
Conventionally, computer-aided diagnosis methods applied to microscopic images of blood smears have used handcrafted features that demand expertise in processing techniques. These are designed to capture morphological, textural and positional variations of the region of interest. Convolutional Neural Networks (CNN) based Deep Learning (DL) models promise superior results with end-to-end feature extraction and classification and could serve as an effective diagnostic aid. This talk will describe our work using DL models for malaria parasite detection in thin blood smear images. We cross-validated the performance of customized and pre-trained CNN models toward classifying parasitized and uninfected red blood cells (RBCs) for improved disease screening. The contributions of this work include: (a) evaluating the performance of pre-trained models as feature extractors, (b) evaluating the untrained architectural frameworks of state-of-the-art DL models, and (c) analyzing the presence/absence of a statistically significant difference in the models’ performance. The analysis was done using RBCs that were mixed across patients as well as keeping them segregated, so that no RBCs segmented from an individual patient’s smears were used in both training and testing.
Dr. Siva Rajaraman is a postdoctoral fellow at the Communications Engineering Branch, LHNCBC. He received his doctorate degree in 2015, in Information and Communication engineering from Anna University, Chennai, India. Before joining NLM, he was an Associate Professor at SSN College of Engineering, Chennai, India. His areas of interest include medical image processing and biomedical signal analysis.