So we are starting to see the AI being used in clinical practice. Like we have AI-based diagnostic tools like Artera AI, which they have the diagnostic and prognostic model for prostate cancer. So there are these new models that can provide more accurate diagnoses, stratify patients for treatment based on treatment response. And I guess we’re going to see more of these models, not only make better predictions, but also to reduce the cost of healthcare...
So we are starting to see the AI being used in clinical practice. Like we have AI-based diagnostic tools like Artera AI, which they have the diagnostic and prognostic model for prostate cancer. So there are these new models that can provide more accurate diagnoses, stratify patients for treatment based on treatment response. And I guess we’re going to see more of these models, not only make better predictions, but also to reduce the cost of healthcare. So we can have AI models that can provide like digital biomarkers that can replace expensive gene sequencing assays. So we can have reduced costs of treatment. We can better screen patients for clinical trials. And the next direction would be where AI provides new insights, things that we don’t do as pathologists or as clinicians, like how to say which patient should go to which therapy to see how the patient’s going to respond, to build the digital twins and see how they will behave under different types of therapies and then we can pick the best one for him, like having this in silico trial for each specific patient. Okay, so another interesting or important direction with AI is improving accuracy and safety of these AI models because now, as I mentioned, we have highly specialized models that do one thing and they do it very well. But we assume that this model is always going to be performing on the type of data it was trained on. But in medicine, there is nothing regular. Every patient is different. There is a lot of variability. And we need these like guardrails that allow the model to make predictions when the data fit its domain expertise and abstain from predictions when it meets some new task or it’s not sure or it cannot just make reliable predictions. So similar to pathologists or clinical fellows, would do. If you are sure about a diagnosis, you know this is like a typical case. You make your sign-out case, you make the diagnosis. If you are not sure, you may go to the tumor board, you may order additional tests. So we want AI models to mimic this behavior, to admit when they cannot make reliable predictions, maybe request additional data, work with other models or other experts to make better predictions. So the important point is increasing safety and accuracy of the AI model when we speak about the translation to clinical practice.
This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.
