A biomarker is a measurable indicator of the changes within the human body, and can serve as a basis to estimate the progression and prognosis of a disease.
‍

Some good examples of biomarkers might be the blood sugar level of a diabetes patient, and the blood pressure of a heart disease patient. A biomarker called PD-L1 can help predict the body’s response to immunotherapy.

‍

‍

PD-L1 is a type of protein created by cancer cells to avoid being attacked by immunocytes. It’s like a mask that cancer wears so the immune system doesn’t recognize it as a threat.
‍

PD-L1 inhibitor is an immunotherapy agent that unmasks and prevents this process, allowing the immune system to act against cancer. Generally, the more PD-L1 is expressed in the body, the better the effect of PD-L1 inhibitors will be.

‍Therefore the rate of PD-L1 expression (the ratio of cancer cells expressing PD-L1 compared to total number of cancer cells) can be a biomarker to predict the effect of PD-L1 inhibitors.
‍

For a stage-4 lung cancer patient, if PD-L1 expression rate is more than 50%, immunotherapy is used. If the expression rate is within 1~49%, the patient can be treated with immunotherapy or combination therapy. If the rate is less than 1%, immunotherapy would have little to no effect, and the treatment will center around cytotoxic agents.

‍

‍

The problem here is that measurement of PD-L1 expression tends to be not very accurate.

‍The current method, done by human pathologists using a microscope to count the expressing cells one by one, can’t help but be inaccurate. Sifting through a slide of cancer tissue is like trying to find a needle in a haystack.
‍

As a result, some patients might get overlooked despite actually being responsive to PD-L1 inhibitors, and some patients might be treated with it but have no effect.
‍

Could AI better find appropriate patients for immunotherapy?