What contributes to burnout in Radiology?

[90sportsfan]

Theres going to be some interesting lawsuits some day. AI says something is suspicious, radiologist is of the opinion it's not and overwrites the report so no further workup is done. Turns out years later, it was real. If AI is known to be better accuracy than humans at that point, was that malpractice? I'd certainly feel like it was if it was my scan it happened to.

Yeah, that's an interesting one. I feel like in scenarios like that, most radiologists would ere on the side of caution, so if AI indicated something, they would definitely make mention of it in the report. I feel like the more common scenario is if AI gave a clean scan, the radiologist would do the read just to double-check and add any additional, potential things that looked suspicious. So basically, you would thrown in any "potential" findings into the report, whether those were discovered by AI or the physicians (just to be safe).

---

Members don't see this ad.

 

[Cognovi]

Yeah, that's an interesting one. I feel like in scenarios like that, most radiologists would ere on the side of caution, so if AI indicated something, they would definitely make mention of it in the report. I feel like the more common scenario is if AI gave a clean scan, the radiologist would do the read just to double-check and add any additional, potential things that looked suspicious. So basically, you would thrown in any "potential" findings into the report, whether those were discovered by AI or the physicians (just to be safe).

A daily occurrence at my place is that the RAPID software gives an abnormal ASPECTS score to a normal head CT. The score is saved in an image in PACS, which is unfortunate, but most of us do not mention it in the report as we are confident there is no CT evidence of an acute infarct. It would only cause confusion and worry among the referrers and the patient.

 

[efle]

A daily occurrence at my place is that the RAPID software gives an abnormal ASPECTS score to a normal head CT. The score is saved in an image in PACS, which is unfortunate, but most of us do not mention it in the report as we are confident there is no CT evidence of an acute infarct. It would only cause confusion and worry among the referrers and the patient.

If you were the attending signing, wouldn't you have a nagging fear that one day you'll do this to a subtle but real stroke?

 

[Cognovi]

If you were the attending signing, wouldn't you have a nagging fear that one day you'll do this to a subtle but real stroke?

For most clinical scenarios, I would rather live closer to the top left corner of my receiver operating characteristic curve than the top right corner. Stroke on head CT is not an exception.

 

[efle]

For most clinical scenarios, I would rather live closer to the top left corner of my receiver operating characteristic curve than the top right corner. Stroke on head CT is not an exception.
View attachment 337950

For most clinical scenarios, I would rather live closer to the top left corner of my receiver operating characteristic curve than the top right corner. Stroke on head CT is not an exception.
View attachment 337950

Makes sense, but here's what mine would probably look like if I was a private practice attending with validated software telling me it sees something haha

 

[maxxor]

For most clinical scenarios, I would rather live closer to the top left corner of my receiver operating characteristic curve than the top right corner. Stroke on head CT is not an exception.
View attachment 337950

One of our Siemens scanners has some nodule CAD which is about as good as breast CAD.

If I see those images in the study, I either mention the nodules if they are real or will specifically say why they are not nodules (“Siemens CAD candidate nodules reflect scarring/pneumonia/aspiration”).

If I were reading a head CT that RAPID or Viz said was positive, I’d probably do the same: RAPID postprocessing indicates a false positive aspects which is felt to be artifactual. (Note is made I only read neuro in the ER setting and hate every minute of it).

 

[guytakingboards]

One of our Siemens scanners has some nodule CAD which is about as good as breast CAD.

If I see those images in the study, I either mention the nodules if they are real or will specifically say why they are not nodules (“Siemens CAD candidate nodules reflect scarring/pneumonia/aspiration”).

If I were reading a head CT that RAPID or Viz said was positive, I’d probably do the same: RAPID postprocessing indicates a false positive aspects which is felt to be artifactual. (Note is made I only read neuro in the ER setting and hate every minute of it).

It's super common for RAPID to spit out a quantitative perfusion numbers leading to an all-caps "THIS PATIENT MAY BE ELIGIBLE FOR THROMBECTOMY" when there's nothing remotely close to a large-vessel occlusion present. I dunno how everyone is cool with that info just sitting in the study jacket for all time.

 

[Cognovi]

One of our Siemens scanners has some nodule CAD which is about as good as breast CAD.

If I see those images in the study, I either mention the nodules if they are real or will specifically say why they are not nodules (“Siemens CAD candidate nodules reflect scarring/pneumonia/aspiration”).

If I were reading a head CT that RAPID or Viz said was positive, I’d probably do the same: RAPID postprocessing indicates a false positive aspects which is felt to be artifactual. (Note is made I only read neuro in the ER setting and hate every minute of it).

Message to medical students worried about AI: the future is already here and it is boring.

 

[rastko]

Neatly pointed out: AI will only cause confusion in radiology, thus adding to the burnout - not help at all. Look at the ECG - have you ever seen a cardiologist, who would believe what the AI spits out? And bear in mind this is only one line.

 

[Radology]

It won’t cause confusion

it will be ignored just like those ecg interpretations and breast cad are