mSickle: sickle cell identification through gradient evaluation and smartphone microscopy

Abstract

Sickle cell diseases are marked by the occurrence of hemoglobin S, which results in the distortion of Red Blood Cells into crescent or sickle-shaped shapes. Thus the detection of sickle cells forms a crucial part in the prognosis or diagnosis of such diseases. The state-of-the-art approaches for mass detection of Sickle cells use deep learning-based methodologies. But these are often computationally expensive and also accompanied by a costly imaging setup. The existing image processing techniques fail to achieve considerable accuracy to be used for detection. The cheaper image processing-based methods do not show competitive reliability as compared to the existing deep learning counterparts. In this paper we propose mSickle, a novel image processing-based approach that can be easily incorporated into a smartphone along with a low-cost imaging device. The geometric properties of the sickle shape have been leveraged to identify the distortion caused by hemoglobin. The performance has been measured with the images in the erythrocyteIDB database and achieved an accuracy of 92.6% across all images. The setup is robust and has been tested across various devices. The presence of sickle-shaped cells acts as a strong indicator of diseases and can be sent for further medical inspection. Therefore, the high accuracy achieved makes it is an image processing alternative to the existing automated deep learning methods for mass detection of sickle cells.