5 ms·

How do you interpret an image if you don't have an understanding of what is in it? In other words, let's say that a patient has a fever. Chest X-ray was done.

How do you interpret an image if you don't have an understanding of what is in it? In other words, let's say that a patient has a fever. Chest X-ray was done. High-school educated "radiologist" sees the image. Does he understand the anatomy? The variations of anatomy? Pathological manifestations of potential causes of fever? How to exclude image artifact versus include potential sign of pathology? What about findings that are not related to fever but need to be identified, further characterised and further imaging required for follow-up? What about signs of infectious fluid versus non-infectious fluid like blood or extravasated fluid? What about when the ordering physician wants to discuss the findings with the "radiologist"? Will that "radiologist" actually understand anything that the physician is talking about? There is a role for AI assisting in rapid analysis of radiological studies, however radiologists can never be replaced because AI will never perform to the level of or have the same functions as a physician radiologist.

treis 8y ago
>How do you interpret an image if you don't have an understanding of what is in it? Because you're looking to identify patterns and match them to known ones. To use an analogy, I can teach you to identify statues of Hindu Gods without teaching you anything about Hinduism. For example, to identify Ganesh you need to know that he has an elephant head. Knowing that he has an elephant head because Shiva cut his human head off doesn't really help you. >In other words, let's say that a patient has a fever. Chest X-ray was done. >High-school educated "radiologist" sees the image. Does he understand the anatomy? The variations of anatomy? Pathological manifestations of potential causes of fever? How to exclude image artifact versus include potential sign of pathology? What about findings that are not related to fever but need to be identified, further characterised and further imaging required for follow-up? What about signs of infectious fluid versus non-infectious fluid like blood or extravasated fluid? All of the things you mentioned are things that a radiologist interpreting a chest X-ray needs to know. What you need to explain is why someone interpreting a chest X-ray needs to understand Organic Chemistry. And the vein structure of the leg. And the typical development pattern of a child. And the various mental illnesses a person might have. And interpret a knee MRI. >There is a role for AI assisting in rapid analysis of radiological studies, however radiologists can never be replaced because AI will never perform to the level of or have the same functions as a physician radiologist. No, but I could easily take 10 people, train them each 1/10th of what a radiologist studies, and have them perform just as well as 10 radiologists by routing the right stuff to the right person.
- riahi 8y ago
  The average Radiologist does not only interpret a single modality or the same body part day in and day out (Radiograph, CT, MR, ultrasound, nuclear scintigraphy, PET, or mammography). We are prepared by our extensive residency to competently read any and all of these upon graduation. We do not interpret these in isolation; we are constantly comparing to prior studies across modalities, often across overlapping body regions. I completely disagree that barely trained hyperspecialists would be sufficient. While specialization is coming to all of Medicine, we tend to cluster by body region/disease state, rather than by modality. This is because all of the modalities provide complementary information and you must be able to cross reference across the various manifestations of disease. All of medical knowledge is iterative. Unlike programming, you can’t just abstract the low-level programming. The premedical curriculum provides the baseline knowledge to understand pharmacology, which is essential when trying to understand our interventions in physiology and pathophysiology. We need to know both physiology and pathophys when interpreting imaging to know how disease manifests and what is abnormal versus post therapy related change. We have to know how referring clinicians will treat disease, and know the major complications to look out for on imaging. We discuss their treatment plans during tumor boards and need to speak the language of the treating teams so we can tailor our interpretations to be useful. There are no easy shortcuts here. NPs and PAs are a living experiment at a shortcut, but what I see day in and day out is that the people they consult (radiology and pathology) need to know even more clinical medicine to help the inexperienced NP or PA in knowing what to do when something happens that deviates from the protocol. Many many many times I will call with a semi-urgent unexpected finding, and just get silence on the other end of the phone. They don’t know what to do, whereas on weekends or nights when I get residents or attendings, I don’t hear this complete absence of understanding.
  - treis 8y ago
    > I completely disagree that barely trained hyperspecialists would be sufficient. >While specialization is coming to all of Medicine, we tend to cluster by body region/disease state, rather than by modality. These are contradictory statements. If Doctors typically go: No Training -> Trained in everything -> Specialized in body region/disease state You can easily go: No Training -> Trained in body region/disease state -> Specialized in body region/disease state So even if we assume it's not possible to specialize radiologists any further than they already are we can still cut the training time and difficulty. >We have to know how referring clinicians will treat disease, and know the major complications to look out for on imaging. We discuss their treatment plans during tumor boards and need to speak the language of the treating teams so we can tailor our interpretations to be useful. We need to know both physiology and pathophys when interpreting imaging to know how disease manifests and what is abnormal versus post therapy related change. No one is denying that there are things in medicine that are complicated and require very skilled people. What you describe is the end state of a fully educated and experienced doctor. Medicine is more or less the only field that makes you become that before you start working. Can you take a reasonably smart person off the street and have him designing plans to treat tumors in a year? No. Can you take that person and train them to identify normal appearing lungs vs cancerous ones in that time? Probably. And can that person learn on the job and develop the ability to design a treatment plan over the course of 10 years working? For sure. And they wouldn't learn everything taught in Med school. They would pick up only the things relevant to the job they are trying to perform. >All of medical knowledge is iterative. Unlike programming, you can’t just abstract the low-level programming. The premedical curriculum provides the baseline knowledge to understand pharmacology, which is essential when trying to understand our interventions in physiology and pathophysiology. We need to know both physiology and pathophys when interpreting imaging to know how disease manifests and what is abnormal versus post therapy related change. Going back to the comment that kicked this all off. If I handed you a cellular biology final from Med school do you think you'd be able to pass it? What about the USMLE? >There are no easy shortcuts here. NPs and PAs are a living experiment at a shortcut, but what I see day in and day out is that the people they consult (radiology and pathology) need to know even more clinical medicine to help the inexperienced NP or PA in knowing what to do when something happens that deviates from the protocol. Many many many times I will call with a semi-urgent unexpected finding, and just get silence on the other end of the phone. They don’t know what to do, whereas on weekends or nights when I get residents or attendings, I don’t hear this complete absence of understanding. Every other field has the ability to take inexperienced people and make them experienced. There's no reason medicine can't do the same. Your clueless NP and PA should have more experienced people to go to for help when they experience something new. The next time they get that call they wouldn't be as clueless.

treis 8y ago

>How do you interpret an image if you don't have an understanding of what is in it? Because you're looking to identify patterns and match them to known ones. To use an analogy, I can teach you to identify statues of Hindu Gods without teaching you anything about Hinduism. For example, to identify Ganesh you need to know that he has an elephant head. Knowing that he has an elephant head because Shiva cut his human head off doesn't really help you. >In other words, let's say that a patient has a fever. Chest X-ray was done. >High-school educated "radiologist" sees the image. Does he understand the anatomy? The variations of anatomy? Pathological manifestations of potential causes of fever? How to exclude image artifact versus include potential sign of pathology? What about findings that are not related to fever but need to be identified, further characterised and further imaging required for follow-up? What about signs of infectious fluid versus non-infectious fluid like blood or extravasated fluid? All of the things you mentioned are things that a radiologist interpreting a chest X-ray needs to know. What you need to explain is why someone interpreting a chest X-ray needs to understand Organic Chemistry. And the vein structure of the leg. And the typical development pattern of a child. And the various mental illnesses a person might have. And interpret a knee MRI. >There is a role for AI assisting in rapid analysis of radiological studies, however radiologists can never be replaced because AI will never perform to the level of or have the same functions as a physician radiologist. No, but I could easily take 10 people, train them each 1/10th of what a radiologist studies, and have them perform just as well as 10 radiologists by routing the right stuff to the right person.

riahi 8y ago
The average Radiologist does not only interpret a single modality or the same body part day in and day out (Radiograph, CT, MR, ultrasound, nuclear scintigraphy, PET, or mammography). We are prepared by our extensive residency to competently read any and all of these upon graduation. We do not interpret these in isolation; we are constantly comparing to prior studies across modalities, often across overlapping body regions. I completely disagree that barely trained hyperspecialists would be sufficient. While specialization is coming to all of Medicine, we tend to cluster by body region/disease state, rather than by modality. This is because all of the modalities provide complementary information and you must be able to cross reference across the various manifestations of disease. All of medical knowledge is iterative. Unlike programming, you can’t just abstract the low-level programming. The premedical curriculum provides the baseline knowledge to understand pharmacology, which is essential when trying to understand our interventions in physiology and pathophysiology. We need to know both physiology and pathophys when interpreting imaging to know how disease manifests and what is abnormal versus post therapy related change. We have to know how referring clinicians will treat disease, and know the major complications to look out for on imaging. We discuss their treatment plans during tumor boards and need to speak the language of the treating teams so we can tailor our interpretations to be useful. There are no easy shortcuts here. NPs and PAs are a living experiment at a shortcut, but what I see day in and day out is that the people they consult (radiology and pathology) need to know even more clinical medicine to help the inexperienced NP or PA in knowing what to do when something happens that deviates from the protocol. Many many many times I will call with a semi-urgent unexpected finding, and just get silence on the other end of the phone. They don’t know what to do, whereas on weekends or nights when I get residents or attendings, I don’t hear this complete absence of understanding.
- treis 8y ago
  > I completely disagree that barely trained hyperspecialists would be sufficient. >While specialization is coming to all of Medicine, we tend to cluster by body region/disease state, rather than by modality. These are contradictory statements. If Doctors typically go: No Training -> Trained in everything -> Specialized in body region/disease state You can easily go: No Training -> Trained in body region/disease state -> Specialized in body region/disease state So even if we assume it's not possible to specialize radiologists any further than they already are we can still cut the training time and difficulty. >We have to know how referring clinicians will treat disease, and know the major complications to look out for on imaging. We discuss their treatment plans during tumor boards and need to speak the language of the treating teams so we can tailor our interpretations to be useful. We need to know both physiology and pathophys when interpreting imaging to know how disease manifests and what is abnormal versus post therapy related change. No one is denying that there are things in medicine that are complicated and require very skilled people. What you describe is the end state of a fully educated and experienced doctor. Medicine is more or less the only field that makes you become that before you start working. Can you take a reasonably smart person off the street and have him designing plans to treat tumors in a year? No. Can you take that person and train them to identify normal appearing lungs vs cancerous ones in that time? Probably. And can that person learn on the job and develop the ability to design a treatment plan over the course of 10 years working? For sure. And they wouldn't learn everything taught in Med school. They would pick up only the things relevant to the job they are trying to perform. >All of medical knowledge is iterative. Unlike programming, you can’t just abstract the low-level programming. The premedical curriculum provides the baseline knowledge to understand pharmacology, which is essential when trying to understand our interventions in physiology and pathophysiology. We need to know both physiology and pathophys when interpreting imaging to know how disease manifests and what is abnormal versus post therapy related change. Going back to the comment that kicked this all off. If I handed you a cellular biology final from Med school do you think you'd be able to pass it? What about the USMLE? >There are no easy shortcuts here. NPs and PAs are a living experiment at a shortcut, but what I see day in and day out is that the people they consult (radiology and pathology) need to know even more clinical medicine to help the inexperienced NP or PA in knowing what to do when something happens that deviates from the protocol. Many many many times I will call with a semi-urgent unexpected finding, and just get silence on the other end of the phone. They don’t know what to do, whereas on weekends or nights when I get residents or attendings, I don’t hear this complete absence of understanding. Every other field has the ability to take inexperienced people and make them experienced. There's no reason medicine can't do the same. Your clueless NP and PA should have more experienced people to go to for help when they experience something new. The next time they get that call they wouldn't be as clueless.