4 ms·
I remember the paper and the conclusion you say is doubtful. Even when the data was heavily distorted and resolution reduced to insanely big pixels it still cou
by andi999 4y ago
I remember the paper and the conclusion you say is doubtful. Even when the data was heavily distorted and resolution reduced to insanely big pixels it still could do this. So, since magic does not exist, there is probably a hidden cofactor (like xrays done on different machines, one in a hospital in a white neighbourhood, the other in a black one, and the machine giving a technical artifact like colorrange). This is just an example of course, I dont know the cofactor.
Here is the study:https://www.thelancet.com/journals/landig/article/PIIS2589-7500(22)00063-2/fulltext https://www.thelancet.com/journals/landig/article/PIIS2589-7...
When i said insanely i didnt remember correctly, just saw even if reduce to 4x4 pixels they claim a statistical difference. So mind blowingly ludicrous resolution might have been a better term.
- walnutclosefarm 4y agoI agree that there is something odd going on in the race study. However, I've seen analogous "discovery" results that don't have this problem. Scientists at my former employer built a model, e.g., that diagnosed asymptomatic heart failure from ECGs. No electrophysiologist in that world-class cardiac medicine unit can do this, but the algorithm clearly perceives the signal. And this was in fact a true "discovery" in that they didn't set out to diagnose heart failure, but rather to "score" ventricular ejection fraction using ECGs. Training the algorithm revealed a signal (a so far unexplained cofactor to reduced ejection fraction) that predicts reduced ejection fraction, and thus heart failure, prior to grossly measurable reduced ejection fraction.
- andi999 4y agoNice. I think this is very interesting, is it published?
- walnutclosefarm 4y agohttps://www.nature.com/articles/s41591-021-01335-4 https://www.nature.com/articles/s41591-021-01335-4
- andi999 4y agoThanks, will read it. The data unavailability is a bit of a bummer.
- coliveira 4y agoAll these "discoveries" suffer from the problem of inconclusive data. As mentioned in another comment, this can be the result of data that is skewed somehow, for example by neighborhood, by the selection of patients, by the technology used, etc. Validating these results is not an easy task, especially when one talks about healthcare data.
- YeGoblynQueenne 4y agoAnd did this discovery generalise to datasets from other hospitals?
- sumy23 4y agoIf you look at their ROC-AUC, performance increases dramatically as resolution increases. Do you really think it's impossible to encode any information in 16 pixels?
- andi999 4y agoNo of course not, obviously there is information encoded. What i think is impossible is that some very subtle difference in the bone structure which experts cannot see which is determined by the self reported race of the xrayed person is still present in that data.