The most promising early warning tests I've seen in the past few years look at degraded DNA fragments floating in the blood that have been shed by cancers. This will probably be the winning technology for all cancer early diagnoses, though genetic biomarkers will probably fail to catch all cancers.
Circulating tumor DNA has several challenges towards being an effective test. The first is just the relatively low proportion of it in the blood (especially before growth is significant) makes reliable detection fairly difficult.
The bigger issue though will likely be managing the false discovery rate. Nearly everyone will have precancerous growths and benign lesions which may share many genetic biomarkers with potentially malignant growths. Many genetic variants associated with cancers occur years or decades before the development of a tumor. It may be difficult to develop a test that doesn't have a difficult tradeoff between sensitivity and specificity, prior to significant tumor growth. I do think its a promising approach, however, and look forward to see the progress researchers make in the coming years.
Relatedly, circulating tumor DNA is not only useful in diagnosis, but its already proving to be useful in the treatment of cancer. Its much less invasive than surgical biopsy, and as a result, clinicians can track the evolution of the tumor, response to specific treatments, and development of resistance in way that was just never possible before.
One wonders if one could train a neural network to look at the mix of junk in the blood and predict what a person will have died of in five years. At this point I'm sure the idea is science fiction. But seeing how DeepMind's AlphaGo played that game, teasing out moves that are beyond human conception routinely, one begins to suspect that there must be some signal that a sufficiently intelligent device could suss out.
I suspect that one decade we will consider this idea almost impossible, and by the end of the next, it will seem old hat. I just wonder which decade that will end up being.
All completely true. But I can easily imagine a near future in which some populations receive screening and then further investigation if something is found. Of course, that creates additional burden on radiology, which is unfortunate, but hopefully all the breakthroughs in ML will make those resources less expensive.
