An Algorithm for Diagnosing Pneumonia

Pneumonia is a common cause of death; over 220,000 people in the UK receive a diagnosis of pneumonia each year. The disease disproportionately impacts children, older adults and hospitalised patients.

Diagnosis relies on blood tests and chest scans but first, a doctor must suspect pneumonia.

Respiratory diseases, such as pneumonia, influenza and the common cold all have one common symptom – coughing. However, the acoustics of coughs are subtly different depending on the cause.

A team at Hanyang University led by Jin Yong Jeon have developed an algorithm based on room impulse responses using cough sounds that can detect if a patient has pneumonia or not. They used an acoustically-aware, machine learning algorithm to identify the typical cough sounds associated with pneumonia and to filter them out from cough sounds caused by other ailments.

Because every room and recording device is different, they augmented their recordings with room impulse responses, which measure how the acoustics of a space react to different sound frequencies. By combining this data with the recorded cough sounds, their algorithm can work in any environment.

Automatically diagnosing a health condition through information on coughing sounds that occur continuously during daily life can facilitate non-face-to-face treatment and reduce overall medical costs.

One company already has plans to apply this algorithm for remote patient monitoring, and the team is hoping to implement it as an app for simple, user friendly in-home care environments.

Algorithms can also assist in the detection of influenza; an algorithm has been developed to assist in the interpretation of influenza testing results and clinical decision-making during periods when influenza viruses are NOT circulating in the community. Most existing influenza tests have high specificity, so false positive results are more common when influenza activity is low.

It is important to ascertain if the patient has an epidemiological link to someone with known influenza. If influenza is detected, antiviral treatment is recommended as soon as possible to minimise its effects in both short- and long-term situations.

Not to be confused with influenza, the common cold virus can alter how a person’s voice sounds because the vocal cords often become inflamed, which changes their acoustic properties. These changes can be detected by algorithm too so, faking sickness will get harder and taking a day off work by nervously coughing down the phone to your boss might no longer ‘cut it” if your company can tell via algorithm if you really have a cold.

Read more here and here