| 1 |
What is the primary goal of the article according to its introduction?
|
To explore advancements, applications, and challenges of generative AI in medical imaging |
|
the reason why I chose the answer that I did is because the article establishes the clinical need for synthetic data due to privacy constraints and rare diseases.
|
It outlines the three pillars of the article being: advancements, application, and challenges.
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 2 |
How do generative AI models differ from traditional discriminative models in healthcare applications?
|
Generative models produce new data rather than only classify or interpret |
|
Generative models learn the underlying probability distribution of the entire dataset, which allows them to sample from that distribution to create novel examples, whereas discriminative models only learn the boundaries between specific categories.
|
My theory would be how I compare the two as the differences between creation and classification.
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 3 |
What is meant by the term “model as a dataset”?
|
Sharing trained model weights instead of raw data |
|
Think of a dataset being an input to provide evidence in order to predict/generate a "model" for doctors to study. Hence why the answer "Sharing Trained Model Weights Instead Of Raw Data".
|
the term "model as a dataset", referencing from the article, refers to a synthetic data set which can be compared to like a "recipe" vs "the finished product". Think of the dataset as the "ingredient" to the "baked cake" (model). Much like GANs or diffusion models, they "study" a data set of real images until they understand well enough to create a "synthetic" image that look real without violating privacy laws.
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 4 |
Which statement correctly distinguishes physics-informed and statistical models?
|
Physics-informed models incorporate biological or physical principles |
|
|
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 5 |
According to the article, what does the “image generation trilemma” describe?
|
Trade-offs among image diversity, quality, and speed |
|
Doctors generally look for an "ideal model" in order to study without violating privacy laws. However, given that generative models cannot meet all three requirements (diversity, quality, and speed,) this becomes a trilemma between trades-offs, becoming a hard to generate an "ideal" study.
|
It becomes a "trade-off" between the three because of the demand that doctors need for a study. You'll either trade the diversity found in human anatomy for a fast, okay-ish quality model, or you'll trade the speed and wait hours for the diversity needed in order to conduct a thesis.
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 6 |
What is the Human Turing Test used for in medical image synthesis?
|
To assess realism of synthetic medical images by experts |
|
it's the gold standard to evaluating realism. While computers can calculate mathematical questions, those numbers will not be able to capture whether if an image is medically plausible.
|
a human expert will be able to spot anatomy errors while a mathematical formula cannot.
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 7 |
Which of the following is NOT mentioned as a potential benefit of synthetic data in healthcare?
|
Eliminating all medical biases permanently |
|
while it is true that the article discusses how to mitigate or address biases, but it acknowledges that AI is a reflection of the data (real human resources) that we provide. Extreme or absolute language is also a red flag. e.g. "permanently"
|
AI is a reflection of the data (real human resources) that we provide; there is no "absolute" outcome when we talk about generative AI.
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 8 |
What is one major ethical concern associated with generative AI in medical imaging?
|
Data copying and patient reidentification |
|
keep in mind that AI is a reflection of the data (real human resources) that we provide--it is trained on real patients. An ethical breach would be that this is a violation of privacy law.
|
As mentioned, AI learns from real resources.
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 9 |
What regulatory precedent did the article cite for synthetic data technologies?
|
FDA clearance of synthetic MRI as image-processing software |
|
|
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 10 |
What is the main purpose of the article?
|
To compare and evaluate ASCVD risk prediction models in East Asia |
|
I choose this answer because the article's introduction and central illustration highlight the inaccuracy and overestimation of western models for East-Asian patients.
|
Westerners appear to be sicker than East-Asian patients. When using a western model following western standards on East-Asians, there will be an Overestimation in prescribing the East-Asian patient.
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 11 |
Which of the following models was originally developed for a Western population?
|
Framingham Risk Score |
|
the Framingham Model would be an accurate answer as it was entirely built on a non-Asian population.
|
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 12 |
Why might Western-based risk prediction models overestimate ASCVD risk in East Asian populations?
|
East Asians have lower baseline incidence of ASCVD |
|
Keep in mind that East-Asians has a diet that drastically differs from westerners, impacting their overall health and lifestyle. While symptoms between both races maybe be identical (e.g. experiencing strokes), these symptoms shouldn't be the only factors to be focused on as it overlooks their different lifestyles.
|
East-Asians lives the lifestyle that drastically differs from westerners, impacting their overall health and lifestyle. Ever wonder why Asians tend to live longer than westerners?
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 13 |
What is the key advantage of the China-PAR model compared to Western-based models?
|
It was calibrated using national data representing diverse regions in China |
|
it addresses the western bias. Models like Framingham were built on a non-Asian population, unlike the China-PAR as it was calibrated using national data representing diverse regions in China.
|
as mentioned, western models were built on a non-Asian population.
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 14 |
Which of the following variables is not typically included in ASCVD risk prediction models discussed in the article?
|
Genetic ancestry markers |
|
knowing your genetics does not change the risk score enough to convince a doctor to change your treatment plan.
|
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 15 |
What is a major difference between the Suita Score and the Framingham Risk Score?
|
Suita Score was designed for a Japanese population using local epidemiological data |
|
The obvious difference would be how these models were to be calibrated. If a model were to be developed using a dataset of a non-Asian population, using it on an Asian would very much be overestimated, whereas using it on a westerner would be the standard.
|
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 16 |
According to the article, what is a potential benefit of developing East Asia–specific risk models?
|
They improve accuracy and reduce overestimation of risk |
|
|
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 17 |
Which factor was highlighted as influencing ASCVD risk differences among East Asian countries?
|
Cultural and dietary variations, such as salt intake and lifestyle |
|
|
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 18 |
What future direction does the article suggest for improving ASCVD risk prediction?
|
Using multimodal AI-based prediction integrated with regional data |
|
in my opinion, improvement would be more in diversity. A region should have a model of its own, not dependent on other regions which might cause an inaccuracy in diagnosis or an overestimation.
|
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 19 |
Which statement best explains the key difference in how VAEs, GANs, and DDPMs generate medical images according to the figure?
|
DDPMs iteratively remove noise through reverse diffusion rather than using encoder–decoder or discriminator structures. |
|
DDPMs is a step-by-step denoising model, while GANs and VAEs adversarial single-step models.
|
"DDPMs iteratively remove noise through reverse diffusion rather than using encoder–decoder or discriminator structures." sounds more like a step-by-step process rather than a single-step to me.
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 20 |
Which of the following best explains the trend shown in Figure comparing age-standardized and crude CVD mortality rates among East Asian countries?
|
Japan and South Korea show low age-standardized CVD mortality rates because of smaller populations. |
|
|
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|