โจทย์ ปรนัย
What is the primary goal of the article according to its introduction?

The introduction mainly talks about different models of generative artificial intelligence, their advancements over the past years, as well as going over many kinds of AI models that can be used in medical practices are of current. Then, the passage goes on to explain the different things the models can do,. The last paragraph of the introduction talks about how this provides an overview of data in medicine, and it will analyse the different kinds of challenges and applictions, as well as future directions. Thus, the main goal of the article must be to "explore advancements, applications, and challenges of generative AI in medial imaging."

This reasoning can be inferred from reading and analysing the introduction of the article, especially the last paragraph.

โจทย์ ปรนัย
How do generative AI models differ from traditional discriminative models in healthcare applications?

The property that makes generative AI different from traditional discriminative models is that the traditional discriminative models are used to interpret data or used in decision making, while the generative AI can study patterns and create new synthetic data that can be used in medical practice.

This can be inferred from reading and analysing the first paragraph of the introduction, as well as the rest of the article which talks about how generative AI can be used to create new synthetic data.

โจทย์ ปรนัย
What is meant by the term “model as a dataset”?

"Model as a dataset" refers to a concept in data sharing, where the generative models can learn and store patterns of original data and compress it within their weights (parameters), which is the structure of the data. It keeps the key features of the training data. These weights can be sent elsewhere, where other models can use the weights to generate synthetic data (that is similar to the original) on their own. Traditional models would have to rely on using actual images (raw data), but by sharing the parameters, data sharing can be more efficient.

This can be inferred from reading "Synthetic datasets - generative models" paragraphs 1 and 2, where it talks about "model as a dataset." It talks about what it means and how it is different from traditional data sharing and models.

โจทย์ ปรนัย
Which statement correctly distinguishes physics-informed and statistical models?

There are 2 types of generative models for synthetic data: physics-informed and statistical models. The physics-informed models use physicl principles and mathematical equations to generate realistic data rather than using patterns from training data. Statistical models learn patterns and characteristics from training data to create synthetic data. Both models require expertise and domain to ensure that the information used to train the model is accurate.

This can be inferred from reading "Synthetic datasets - generative models", specifically the paragraphs containing information about physics-informed and statistical models, where by analysing how each model works, the similarities and differences can be seen.

โจทย์ ปรนัย
According to the article, what does the “image generation trilemma” describe?

The article mentioned 3 kinds of statistical models: VAEs, GANs, and DDPMs. Each of the models have their own strengths and weaknesses when it comes to generating synthetic data, which can include the quality of the data, how much coverage a dataset may have, and how fast the sampling rates are. It is a "trilemma" because three of these properties are important for generating synthetic data, but there may be a trade-off depending on the kind of model used, so there should be a balance, as well as choosing each model based on the goals of the user.

This can be inferred from reading and analysing "Synthetic datasets - generative models", specifically the paragraphs about statistical models and their trilemma encounter. By reading each models' functions, we can understand what the image generation trilemma is.

โจทย์ ปรนัย
What is the Human Turing Test used for in medical image synthesis?

While creating synthetic data, it is important to ensure that the data is close to real data in order to prevent any potential biases or errors if that data is used in real situations. One of the ways to test the quality is the "Human Turing test," where experts are asked to distinguish between what is real data and what was generated. By doing this, insights on the realism and accuracy of the generated data can be improved.

This can be inferred from reading and analysing the "Human evaluation" section, which contains information about the Human Turing test and how it plays a role in synthetic data.

โจทย์ ปรนัย
Which of the following is NOT mentioned as a potential benefit of synthetic data in healthcare?

Although synthetic data may be able to eliminate some bias, it is difficult remove every single bias permanently. Depending on the source dataset, it can lead to potential biases. Training data that is leaning towards a certain demographic, such as race, age, gender, income and such can cause the synthetic data to also lean towards a certain bias. Using this data as a generalisation can be inappropriate towards minorities, making it inaccurate.

This can be inferred from reading and analysing the "Potentials and promises" and the "Challenges and considerations" section. From it, we can understand the benefits and disadvantages of using synthetic data and rule "eliminating all medical biases permanently" out as the answer.

โจทย์ ปรนัย
What is one major ethical concern associated with generative AI in medical imaging?

In terms of ethical concerns in medicine, patient privacy is most important. The generative AI trained on real dataset, for example, MRIs, may keep real patients' facial structure and copy it into synthetic data. The structures may be retraced, making the patients identification no longer private. The other options are mainly about the model in itself, which is a technical/practicality issue rather than an ethical issue.

This can be inferred from reading and analysing the "Challenges and considerations - patient privacy and data copying" section, which talks about how the data may be copied and being able to reidentify the patient, making it a major ethical concern.

โจทย์ ปรนัย
What regulatory precedent did the article cite for synthetic data technologies?

The article cites the FDA to ensure that radiologist performances remained the same while using real and synthetic datasets.

This can be inferred from reading "Future directions", where the article explicitly cites the FDA.

โจทย์ ปรนัย
What is the main purpose of the article?

The article's goal is to study how well the risk prediction models can perform on East Asians between those living in Asia and those living in the USA.

This can be inferred by reading the abstract of the article.

โจทย์ ปรนัย
Which of the following models was originally developed for a Western population?

The Framingham Risk Score was developed by 4 US community cohorts, and a risk factor included race, which hints that it was developed for the entire Western population rather than one specific sub-group.

This can be inferred by reading "Current state of ASCVD risk alculatos for East Asians," where it talks about the Framingham study.

โจทย์ ปรนัย
Why might Western-based risk prediction models overestimate ASCVD risk in East Asian populations?

The Western models use data from their populations, which have a higher baseline incidence of ASCVD compared to Asians (lower 10 year risk), which averages into a higher baseline, inaccurate to East Asians. The risk distribution is different.

This can be inferred by reading "Current state of ASCVD risk alculatos for East Asians," where it talks about how East Asians have a lower risk estimate comared to White people.

โจทย์ ปรนัย
What is the key advantage of the China-PAR model compared to Western-based models?

The China-PAR model uses data from their own country and people, which makes it more accurate to their population because of how the risk factors are distributed more evenly.

This can be inferred by reading "ASCVD risk prediction in China", which talks about the China-PAR model.

โจทย์ ปรนัย
Which of the following variables is not typically included in ASCVD risk prediction models discussed in the article?

Age, blood pressure, smoking status, and cholesterol can contribute to the chances of getting ASCVD, but ancestry markers may not be a key risk in itself.

This can be inferred by reading about the different kinds of risks that can contribute to ASCVD from the article.

โจทย์ ปรนัย
What is a major difference between the Suita Score and the Framingham Risk Score?

The Suita score was chosen from 10 risk prediction scores in Japan, making it more accurate for the Japanese population, while the Framingham includes mainly the Western population.

This can be inferred by reading "ASCVD risk prediction in Japan", which talks about the Suita score.

โจทย์ ปรนัย
According to the article, what is a potential benefit of developing East Asia–specific risk models?

Western models such as the Framingham model uses cohorts of the Western people, which causes it to be inaccurate and overestimate risks for East Asians. By developing models specifically for the East Asian population, it can be more accurate to describe East Asians and warn them of risks.

This can be inferred from reading about the Framingham risk score and its inaccuracy when used on East Asians, as well as reading how each EA country developed their own models with local data to improve accuracy.

โจทย์ ปรนัย
Which factor was highlighted as influencing ASCVD risk differences among East Asian countries?

Within East Asians, the baseline and other factors of ASCVD are more common with each other compared to Western populations. The key difference is the diets and lifestyle of each individual, sub-group, or country as a whole.

This can be inferred by reading the sections is which the article highlights the need of studying cultural differences to get a more accurate data representation.

โจทย์ ปรนัย
What future direction does the article suggest for improving ASCVD risk prediction?

By using regional data, the risk prediction and scores can be more tailored to their populations. Using a multimodal AI-based prediction provides a path for using multiple biomarkers to identify the health of each patient, each makes it able to cover more factors. Overall, it can be more accurate.

This can be inferred by reading the future directions and conclusions of the article, which talk about how the ASCVD risk prediction models can be improved.

โจทย์ ปรนัย
Which statement best explains the key difference in how VAEs, GANs, and DDPMs generate medical images according to the figure?

VAEs work by compressing data then reconstructing it to generate new, similar data. GANs work by using dual-systems: generators create the data while discriminators evaluates it and provides feedback as to what might be real or fake DDPMs work by denoising then estimates te data distributon nd refines te samples repeatedly to create high-quality data. DDPMs do not use encoder-decoder or discriminator structures.

This can be inferred by reading about the 3 statistical models provided by the article to see how each functions, as well as looking at the provided image.

โจทย์ ปรนัย
Which of the following best explains the trend shown in Figure comparing age-standardized and crude CVD mortality rates among East Asian countries?

China's age-standardised rate is 277 while the crude mortality rate of CVD is 322, which is a rather big difference, indicating an overstimation or underestimation.

The answer is inferred by looking and analysing the 2 provided charts.