BY DR. PARAG BHAMARE
Cough is an important symptom of respiratory illness but it’s very non-specific. The value of cough as an indicator of overall health of the respiratory system is unquestioned but there is limited objectivity to this analysis. Features of cough such as onset, duration, progress and other associated characteristics are utilised by clinicians. These are essential enquiries while taking history and they add an important insight for the differential diagnosis. However, use of cough sounds or in other terms acoustics of cough is not quite widely used, except in certain cases where we get specific characteristic sound, e.g. high-pitched whistling sound in asthma.
In general, doctors, especially senior clinicians train their ears to recognise the acoustics of cough. Many use it either subconsciously and largely attribute this ability to differentiate between coughs to their overall experience and assessment of the patient.
Let’s take another example. In families where a child suffers from asthma, people in the household learn to identify when a child is coughing because of, say, common cold, or if it is a prelude to asthmatic attack. Ask, how they know, and the answers are vague. It is because this kind of “expertise” is difficult to explain and pass on. One of the reasons is that ears have to be trained over time to catch small changes. But that isn’t an exact science. It is a guesstimate. Subtle changes in the sound are difficult for the human ear to capture but a computer can possibly do it well. It can track the smallest change in the waveform. This makes me believe the acoustics of cough sounds can be assessed and utilised.
Can this be applied to early detection of COVID-19? Or can we think of using early signals from cough sounds to arrive at a subset of COVID-19 suspects? Can we use this to identify those who might progress to severe forms of disease and catch impending lung involvement early in the disease progression? I argue that it is possible.
I need to be honest with you. This is a moonshot project. But there is some anecdotal evidence which convinces my colleagues at Wadhwani Institute for Artificial Intelligence (Wadhwani AI) to believe that there is a possibility of this moonshot. The researchers at Wadhwani AI argue that a model, which detects, and analyses cough sounds could be the answer to early detection of COVID-19. It is a gamble, but given the situation, it is worth rolling the dice.
Let’s look at this purely through the lens of medicine.
Different respiratory ailments caused by viruses, bacteria or other pathogens have differences in clinical, radiological and pathological manifestations. These diseases do have differential involvement of lungs, airways and variable immune responses. If we think of cough sound; irritation or involvement of lung tissue or airways, the lumen of airways, fluidity and consistency of secretions are some of the most important factors that largely determine the nature and acoustics of sound. For instance, viral onset secretions are white in colour, watery in consistency, bacterial ones are yellow with a stickier character. Different pathogens also have the propensity to affect different parts of the lung or respiratory tract. As per available medical evidence, in COVID-19, dry cough is one of the early presenting symptoms with or without fever. With progressing infection, the patient might develop cough with expectoration, shortness of breath and eventually, if not managed in time, may suffer from respiratory distress, pneumonia and other severe forms of disease.
The basic assumption here is to see if these differences in infected area, fluidity of secretions and/or condition of airways reflect on the acoustics of cough sounds and can we utilise the strength of artificial intelligence to get early signals about this progression or possibly disease itself in the first place.
The acoustics of cough can be considered as an objective addition to overall patient assessment. Cough sound acoustics coupled with vital parameters such as temperature, pulse and respiratory rate can add further strength to decision making. In any viral infection, and more so in Severe Acute Respiratory Syndromes (SARS) if the patient’s condition is to worsen, the negative progression is rapid. Any intervention, which could help us to catch it early and slow down the infection is welcome.
This becomes more important when we have limited testing capacities, and most patients have mild symptoms (limited to fever and/or cough). A possible standard tool to get symptom positive patients screened for COVID-19 would not only help an overburdened healthcare system prioritize patients but also avoid preventable deaths.
I would urge you to participate in our global data-crowdsourcing and open-innovation initiative here.