Tuberculosis is the second leading cause of death from infectious disease in developing countries. Its high mortality rate is due to a variety of factors, including the lack of available and affordable diagnoses and inconsistent results acquired in patient follow-up. Tuberculosis is controlled by taking a sputum sample and manually counting the cells under a microscope. In low-resource countries, this is very difficult, as there are few qualified personnel working under difficult conditions. Doctors may also need to check ten patients per day, and for each patient hundreds of cells need to be counted through a microscope. This leads to eyestrain for clinicians and slow, poor-quality results for patients.
To put their grain of sand against this disease, Cambridge Consultants, an expert in the development of innovative technological solutions, has developed BacillAI, a smartphone that uses artificial intelligence (AI) to improve the monitoring of tuberculosis treatment in countries with resources limited. The phone captures the images from the laboratory and analyzes them using a deep learning algorithm to identify, count and classify tuberculosis cells, in order to determine the patient’s disease status and the appropriate treatment for each case.
How does BacillAI work? The stained sputum sample images are analyzed using the smartphone and a deep learning algorithm, specifically a convolutional neural network, to identify, count and classify Tuberculosis cells, in order to determine the patient’s disease status. The test results are returned to the doctor through a dedicated app. An automated system for counting cells and classifying treatment progression offers a variety of benefits, including increased consistency and throughput, and automatic digitization of results.
The development of BacillAI is a multidisciplinary project that has required an expert team to develop a high-performance artificial intelligence system that works with suboptimal or low-contrast images, and that can be implemented on devices with limited computing power. An imaging and support system for the smart phone has also had to be designed, which easily aligns the smartphone camera with the microscope’s optics to transmit good images.
Artificial intelligence training
In addition, artificial sputum had to be developed and bacterial cells cultured to create the large number of slides needed to support AI training, and bundled the system for the user with an intuitive and easy-to-use application interface, with a easy to assemble hardware.
Cambridge Consultants works at the forefront of advances in artificial intelligence applying deep learning in applications including anomaly detection in telecommunications networks, autonomous cars, frictionless user interfaces, and medical diagnosis and treatment. Identification of tuberculosis cells is particularly difficult due to the presence of other similar-looking cells in sputum samples, the variation in the intensity of the spots, and the variation in color between imaging devices. These conditions cause traditional rule-defined computer vision techniques to fail as they cannot account for such breadth of data. The rich and varied data set provided by BacillAI creates the ideal conditions to train a robust and accurate deep learning system for cell identification.
Dr Kathleen England, Senior Advisor for Tuberculosis Diagnostics at Cambridge Consultants, and formerly Doctors Without Borders, notes that ” a system that eliminates human variability in counting, lowers skill level, increases performance, and shares information through of a network is very beneficial for the examination of the patient’s smear . “
BacillAI has been developed using Cambridge Consultants specially designed deep learning research facilities, implementing NVIDIA’s DGX POD and Netapp ONTAP AI architectures, which provide high-performance flash storage with the capacity for several petabytes of information. .
