How Artificial Intelligence Will Change Lung Disease Diagnosis, Part 1: Where We Stand.

According to WHO, 3 out of 5 top leading causes of death in the world relate to a lung dysfunction. Pneumonia is the leading cause of death in children under 5 during postneonatal period, accounting for 13%. Over 317 million people find it really hard to breathe because of chronic respiratory conditions (64 million have COPD, 253 million are coping with asthma).

Intimidating, right? The statistics certainly alarmed us, and urged to dig deeper in search for answers. What are the reasons for such high numbers? Can we protect ourselves and those close to us or should we just accept the situation as is? Here is what our team has found.

The ‘What’

We’ve been told time and time again to be mindful of our health and to actually care for our bodies. But it’s so easy to dismiss a prolonged cough or a shortness of breath when you’ve got this new vibrant opportunity to seize that just won’t wait another day.

After all, you can google the symptoms, discard the most disturbing cases because ‘it can’t happen to me’ and decide that you’ll get better eventually, right?

Well, no. Here is an example. Idiopathic pulmonary fibrosis (IPF), a fatal disease mostly affecting people over 50, has a median survival of only 3 years. If a person is diagnosed at the early stage, existing therapy will be more efficient, granting him a longer and better life. However, the average delay between first symptoms and referral to an IPF centre is 2.2 years! At first, there’s the reluctance from the patient side to go to the doctor. But there are more obstacles on the way to recovery.

While lung auscultation is a core part of regular check-ups, an average doctor’s ability to correctly identify lung sounds is only poor to moderate. In a study published in Australian Journal of Physiotherapy, the accuracy rates ranged from 12.5 to 62.5%. The results of reliability evaluation were even worse.

After listening to the same lung sound recording mixed in between others, only 53% of doctors correctly identified the sounds from that recording twice and 17% managed to do it three times.

Unsurprisingly, it gets substantially harder to treat a condition efficiently when you can’t reliably tell what the condition is.

So there it is, a triad that will lead you nowhere good: postponing check-ups, ignoring symptoms and relying on the possibilities of human hearing. The first two are in your hands, but is there a way out of the diagnostic trouble? We are thrilled to testify — yes, there is! A hint: the answer lies in the field of artificial intelligence. Stay tuned to know about the solution in our next article.