Oliver S. Zhao, Nikhil Kolluri, Annie Anand, Nicholas Chu, Ravali Bhavaraju, Aditya Ohja, Sandhya Tiku, Dat Nguyen, Ryan Chen, Adriane Morales, Deepti Valliappan, Juhi Patel, Kevin Nguyen
Malaria is an infectious disease caused by Plasmodium parasites, transmitted through mosquito bites. Symptoms include fever, headache, and vomiting, and in severe cases, seizures and coma. The World Health Organization reports that there were 228 million cases and 405,000 deaths in 2018, with 93 and 94 percent of total malaria cases and deaths occurring in Africa, respectively. Rapid diagnosis and subsequent treatment are the most effective means to mitigate the progression into serious symptoms. However, many fatal cases have been attributed to poor access to healthcare resources. In these low-resource settings, the use of light microscopy on a thin blood smear with Giemsa stain is used to examine the severity of infection, requiring tedious and manual counting by a trained technician.
To address the malaria endemic in Africa and its coexisting socioeconomic constraints, we propose an automated, mobile phone-based, screening process that takes advantage of already existing resources. We utilize an SSD-based object detection architecture that rapidly processes thin blood smears acquired via light microscopy to isolate images of individual red blood cells. Then we implement a FSRCNN model that upscales 32×32 low-resolution images to 128×128 high-resolution images with a PSNR of 30.79, compared to a baseline PSNR of 24.10 through traditional bicubic interpolation. Lastly, we utilize a modified VGG16 model that classifies red blood cells as either parasitized or uninfected with an accuracy of 96.5% and AUC of 0.9940. Our deep learning platform is efficient enough to operate exclusively on low-tier smartphone hardware, reducing the need for cloud-based computing.
These sequential models on streamlined platform give healthcare providers the number of malaria-infected red blood cells in a given thin blood smear sample. The use of an automated screening process with only basic resources can allow for significantly greater screening measures in low-resource communities.
Comments
This is excellent — you clearly spell out the problem, the steps taken as part of your solution, and you illustrate how it plays out on a mobile platform. You’ve really used the design of the poster to walk the viewer through all these steps. Great job! —Rob Reichle
Thank you! We appreciate the encouragement! —Oliver S. Zhao
This is a beautifully-designed poster, and such an interesting project! The numbers on malaria are so stark, and the research seems very promising for better diagnostic capabilities. What is the next step for this project? Are you still working on it? —Jeanette Herman
Thank you for the encouragement Jeanette! Indeed the numbers are very stark. When we first began this project, I distinctly remember checking the World Health Organization report several times, just because 228 million cases seemed like such a large number that I thought surely it was a mistake, but sadly it was not. We are still working on the project! There are several more obstacles to overcome, but perhaps the most immediate obstacle is working on improving the SSD object detection model, which we have only began developing several weeks ago. As you may have already noticed on the poster, the average recall of the object detection model is quite poor, at only 63.9%. As in any sequential process, the weakest link defines the strength of the chain, and right now our weakest link is the object detection model. —Oliver S. Zhao