@rubin - thanks for writing this blog post. I wish you had published this as a peer reviewed opinion paper so that you would have gotten feedback from reviewers and would have improved this piece.
I absolutely agree that AI can significantly impact Radiologists' interpretation times, patient outcomes and also ROI.
My issue with this blog is that you are using interpretation times that are not realistic. As someone who has done dozens of reader studies in his life, average xray interpretation time is actually close to 90 seconds and not 15 mins. If you look at just Chest Xrays, it is much lower. Here is a recent Nature peer reviewed paper (https://www.nature.com/articles/s41746-023-00829-4) showing Chest Xray interpretation time of 15 seconds without AI and 13 seconds with AI. I think you should redo the calculations using more realistic interpretation times.
Here is a paper that actually excluded cases that took longer than 120s. (https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0264383) Here is the reason why: To exclude the cases that remained open for long durations due to unexpected interruptions, we considered more than 120s as an unreliable reading time because readers may have been interrupted and excluded from the analysis.). 15 min interpretation time is totally unrealistic. In this study, the interpretation time actually went up from 14s without AI to 19s with AI.
I hope you will consider this as a positive criticism of your post.
@rubin - thanks for writing this blog post. I wish you had published this as a peer reviewed opinion paper so that you would have gotten feedback from reviewers and would have improved this piece.
I absolutely agree that AI can significantly impact Radiologists' interpretation times, patient outcomes and also ROI.
My issue with this blog is that you are using interpretation times that are not realistic. As someone who has done dozens of reader studies in his life, average xray interpretation time is actually close to 90 seconds and not 15 mins. If you look at just Chest Xrays, it is much lower. Here is a recent Nature peer reviewed paper (https://www.nature.com/articles/s41746-023-00829-4) showing Chest Xray interpretation time of 15 seconds without AI and 13 seconds with AI. I think you should redo the calculations using more realistic interpretation times.
Here is another peer-reviewed study showing reduction of reading times with AI from 10–65 seconds to 6–27 seconds (https://pubs.rsna.org/doi/10.1148/radiol.2021202818).
Here is a paper that actually excluded cases that took longer than 120s. (https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0264383) Here is the reason why: To exclude the cases that remained open for long durations due to unexpected interruptions, we considered more than 120s as an unreliable reading time because readers may have been interrupted and excluded from the analysis.). 15 min interpretation time is totally unrealistic. In this study, the interpretation time actually went up from 14s without AI to 19s with AI.
I hope you will consider this as a positive criticism of your post.