Machine learning could be used for simpler way to diagnose pneumonia infection in children

Machine learning could be used for simpler way to diagnose pneumonia infection in children

Researchers at Oxford University are developing a tool to make it easier and cheaper to diagnose pneumonia, the number one killer of children under five. Their latest research is published in the Royal Society Interface Journal. Currently, properly diagnosing pneumonia and understanding its severity requires specialized doctors and expensive equipment, such as X-ray machines. Neither approach is available to community health workers in developing countries, where 99% of the 1.1 million childhood pneumonia deaths occur each year.

Elina Naydenova from Oxford University’s Institute of Biomedical Engineering explained: “Generalist health workers rely on a set of guidelines known as the IMCI during the working hours closest to the hospital. These guidelines can sometimes Pneumonia cases are well identified, but not well screened There is also considerable variability among users in different settings. Unnecessary antibiotic prescribing in settings where no clinical expert can make a conclusive diagnosis The numbers have increased, thereby depleting vital drug supplies and increasing the problem of antibiotic-resistant infections. We hope to apply smart engineering techniques to develop robust automated systems that are consistently more accurate.”

Accurate diagnosis can reduce mortality by 42%, but not only correctly identifies children with pneumonia. Health workers also need to judge the severity of the infection to determine whether the child needs to be referred to a hospital, and whether the infection is bacterial or viral to determine if antibiotics are making a difference. However, they just need to be able to do all of this with a basic set of portable equipment.


For an automated system to be effective, it must be able to use data from this basic device, so the Oxford team took in-depth data from a clinical study in The Gambia and used machine learning techniques to see if they could develop a diagnostic tool that could diagnose Algorithms for pneumonia.

Elina said: “To identify pneumonia, we found four characteristics that can be measured with two devices. Heart rate, respiratory rate and oxygen saturation can all be measured using a pulse oximeter. Temperature requires a thermometer. These are all things that can be made Available to health workers with basic training. “Using these four measures compared to IMCI, we achieved a sensitivity of 98.2% and a specificity of 97.5% (i.e., they could correctly identify every 1000 cases 982 of pneumonia cases, while falsely identifying pneumonia in 25 per 1,000 people without the disease]the best performance was 94% sensitivity and 69% specificity.”

Using a stethoscope to assess both lung sounds, the team was able to determine the severity of the infection with a sensitivity of 72.4% and a specificity of 82.2% (IMCI reached 79.3% and 67.7%, respectively). While the team noted that adding the biomarker C-reactive protein (CRP) to the test provided 89.1 percent sensitivity and 81.3 percent specificity, the team noted that this would involve additional costs.

Finally, by simultaneously assessing heart and respiratory rates and blood oxygen saturation using a biomarker called Lipocalin-2, the team could identify whether pneumonia was bacterial or viral with 81.8 percent sensitivity and 90.6 percent specificity. When IMCI is used, it is 100% sensitive to serious bacterial infections but 0% sensitive to specific bacteria – all severe viral cases should also be prescribed antibiotics that don’t cause any effects. While low-cost tests for these biomarkers are not yet commercially available, many research groups around the world are already investigating the development of such tests for use in resource-constrained settings.

Elina said: “We have identified a set of functions that can replace the combined X-ray and blood culture only available in fully equipped hospitals. These functions will be used in a mobile app linked to a suite of low-cost products Diagnostic equipment, which we will try out for years to come.”

The Links:   TM121SV-02L07D MG200J2YS1