Medical Devices and Machine Learning

The best medical products were developed in the West. Inevitably, clinical trials were done on Western folks. Were there differences between white people and the rest (black, brown, yellow) that were relevant to the product – nobody knew for sure. The trust in Western processes, the aura around their products, and the lack of alternatives meant the rest of the world would use those medical products.

 

But now, writes Rahul Matthan, concerns are being raised. He cites pulse oximeters, those easy-to-use devices that measure oxygen saturation. Their usage spiked as even the common man started using them as a quick and cheap check to be run during COVID-19 times.

“The fact is that these devices have largely been tested on lighter-skinned people, their algorithms tuned to the light absorption and reflection characteristics of paler skin. As a result, they perform poorly on darker complexions.”

The same problem, he says, is being found with melanoma (cancer) detection algorithms. The algorithms used in ECG machines are being found to have the same problem.

 

The problem, he says, is hard to fix. Those algorithms are proprietary and often hidden behind patents, so others can neither view them nor evaluate them for (unintended) bias and blind spots. Even when doctors suspect a problem, they have no clue how to compensate for “what their devices are telling them”.

 

He fears that with AI and machine learning, the dangers will be even higher. Even the inventors won’t understand how the algorithm works, let alone its mistakes.

 

One solution, he says:

“(A possibility is) releasing the underlying algorithms as open source so that researchers, software engineers and medical practitioners alike can analyse them, detecting any bias that may exist and, where possible, modifying them to better suit the patient populations they are being used on.”

But he admits this is easier said than done – companies wouldn’t be keen to open up their algorithms and data sets.

 

I wonder though, whether the West and countries like China and India will start to diverge on this front. The Western system is based on the idea of trade secrets and patents. Neither China nor India cares much about these. Hence, thinking of (and building) a different system, keeping AI and machine learning in mind, would not be disruptive to existing companies in these countries. Nor would there be a pushback and lobbying against it as would be the case in the West. We can already see such divergences happening with digital payment systems catching on like wildfire in India and China, while they barely make a dent in the West. In all matters digital, the West is running into the problem of legacy (existing) systems, while the East is unburdened by such things.

Comments

Popular posts from this blog

Student of the Year

Animal Senses #7: Touch and Remote Touch

The Retort of the "Luxury Person"