We have developed a revolutionary new framework for telepathology, creating a new paradigm for remote medical image diagnosis. This platform relies on the participation of non-expert crowds of individuals by playing bio-games that are designed around specific pathology tasks. To demonstrate the capabilities of this platform, we have designed a game for distributed diagnosis of malaria. Malaria is a major health problem in many tropical and sub-tropical climates, including much of sub-Saharan Africa. It is the cause of ~20% of all childhood deaths in this region, and almost 40% of all hospitalizations in whole of Africa. For diagnosis of malaria, conventional light microscopy remains the gold standard method. A pathologist must typically check between 100 and 300 different field-of-views of a thin blood smear (corresponding to inspection of at least 1,000 individual red blood cells) using a light microscope with 100X objective lens before being able to reliably call a thin smear sample negative (i.e. not infected). This is a very time-consuming task and a significant challenge given the large number of cases observed in these resource-poor settings. Approximately 60% of the cases reported in sub-Saharan Africa are actually false-positives, and any reduction of such cases can reduce unnecessary treatments and hospitalizations. We have demonstrated that it is possible to approach the accuracy of medical experts in making cell-level diagnosis of malaria by collecting and combining the responses of non-expert gamers. Specifically, we have reported an achieved diagnostic accuracy which is within 1.25% of a trained medical professional with only 20 gamers. The details can be found in our latest publication. The designed game can be played on a variety of platforms. You can play it with your browser on any Flash-capable system or any Android-based device, such as a mobile phone or a tablet.
Manuscript in PLoS ONE: PDF
Manuscript in Lab on a Chip: PDF
Manuscript in Games For Health: PDF
Manuscript in PLoS ONE: PDFFigure 1 The game can be played on multiple platforms, including Android phones and tablets.
As more and more individuals play this game, their responses are combined using an intelligent algorithm to achieve evermore accurate results. In doing so, the algorithm treats the overall process as a noisy telecommunication system. The images are treated as information that is broadcast simultaneously to multiple gamers. Each individual gamer then acts as a noisy repeater that transmits back his diagnoses. The gamer responses are then used to decode the most likely image labels for the original images. This decoder takes a Maximum a posterior Probability (MAP) estimation approach in computing the most accurate diagnostic decisions - a methodology borrowed from traditional communications theory.Figure 2 The gaming system can be described as a telecommunication network where the images are broadcast to a set of repeaters that output their noisy diagnoses. These diagnostic responses are received by a decoder that combines all the gamer responses and outputs the most likely labels for the original images.
So far we have shown that this platform is capable of achieving high accuracies in diagnosing red blood cells that are potentially infected with malaria. We have shown this on a small scale with up to 30 gamers. We need your help to scale this up into a truly massively crowd-sourced platform. When you play a game, your responses are collected and combined with those of other individuals to produce an accurate overall diagnosis. Our goal is to achieve the same accuracy level of a medical professional. Having shown that the crowd's accuracy increases with the size of the crowd, we are interested in finding the most optimal number of individuals needed for accurate diagnosis.Figure 3 The envisioned micro-analysis network
Over the past five years our research team has been working on developing low-cost, compact microscopy and micro-analysis devices including ultra-portable holographic microscopes, mobile phone based microscopes, and mobile phone based flow cytometers (See Figures 4). More recently we have developed a mobile phone-based Rapid Diagnostics Test (RDT) reader (See Figure 5). Our vision and long-term goal is to create a self-learning integrated network of microscopes toward smart biomedical micro-analysis and diagnosis. We can achieve this through portable hardware that facilitates low-cost, high-throughput imaging of biological samples along with software and intelligent algorithms that can analyze such data through a crowd-sourced telepathology platform.