On 2017 Oct 15, Nevit Dilmen commented:

Let's clarify the hypothesis about the potential use of AI in mammography:

Some cancers which are present in the patient's breast have a footprint on mammography but the human radiologists are unable to detect the sign in some cases.

AI might help in improving above category.

AI can NOT help if a cancer has no footprint on a mammogram. Indeed mammograms do not show cancers in some cases and we should not expect any benefit of using AI in those cases. Examples are any tumor that is too small, any tumor outside of the imaged area, dense breasts.

Much of the effort to improve screening is focused on reading. AI might also help to clarify how much we should expect from mammography itself as a modality. How much human readers are actually missing.

On 2017 Oct 15, Nevit Dilmen commented:

Let's clarify the hypothesis about the potential use of AI in mammography:

Some cancers which are present in the patient's breast have a footprint on mammography but the human radiologists are unable to detect the sign in some cases.

AI might help in improving above category.

AI can NOT help if a cancer has no footprint on a mammogram. Indeed mammograms do not show cancers in some cases and we should not expect any benefit of using AI in those cases. Examples are any tumor that is too small, any tumor outside of the imaged area, dense breasts.

Much of the effort to improve screening is focused on reading. AI might also help to clarify how much we should expect from mammography itself as a modality. How much human readers are actually missing.