Researcher here. I just read this piece about medical imaging ai with object recognition and it left me wondering why there are still 100,000+ deaths a year in the US due to misdiagnosis - anyone out there working on these problems? Vinod Khosla famously said that he'd rather get surgery from AI than from a human - so where are we at with that?
3
-
$\begingroup$ "so where are we at with that?" That's a very, very broad question. I also don't think it's just limited by AI, but by other high-precision mechatronics not being economically viable at this time. $\endgroup$– MastAug 24, 2020 at 19:21
-
1$\begingroup$ Welcome to SE:AI. "Improving faster" is subjective--this question might be better served with historical stats on diagnoses errors. (Short answer--tech takes time to develop & implement, and prove itself in the field. Also, new tech tends to be very expensive initially, especially in the medical sector. We've only had strong-narrow statistical AI for about 5 years now.) $\endgroup$– DukeZhouAug 24, 2020 at 21:41
-
$\begingroup$ Thank you for your help. $\endgroup$– Very AIAug 30, 2020 at 16:27
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
AI in healthcare is already playing a big role in diagnosing the diseases, assisting patients or helping the medical staff in supplying various things or performing actions.
The medical imaging helps AI to diagnosis the diseases without help radiologist. And to develop the AI model using medical images to diagnosis various disease need training datasets for machine learning algorithm helping to detect with accuracy.
But just like AI, medical imaging is not improving faster making sometimes difficult or more time taking to diagnosis the diseases. Actually, while developing such highly sensitive models, high-quality training data is required and there is lack of such annotated data to train the models.
The more training data is feed into the model, it will learn the detection process with more accuracy resulting predicting the faster diagnosis. And to improve the quality and quantity of data, more dedicated and fast labeling process is required.
Hence, data annotation companies now use the AI-assisted labeling process to annotate the medical images at faster speed with better accuracy. Compare to manual annotation, AI companies can get multiple times faster-annotated data for machine learning algorithms.
Once the training data will start available in large quantity for medical imaging analysis process through AI will also improve. The AI-assisted data annotation process can only help to produce the medical imaging datasets in large quantity for better predictions in healthcare sector.
