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

The article's introduction clearly define its scope as a critical analysis of generative AI's advancements, applications, and challenges in medical imaging, not hospital management, economics, policy, or model design.

Stated in the Introduction, found in page 1, of article number 1. The article structures its scope around three main areas are image generation paradigms, practical applications, and ethical challenges, which shows that the article's purpose is to advance understanding of generative AI in medical imaging, directly supporting the correct answer and ruling out all other options that fall outside this defined scope.

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

The article states, "Generative artificial intelligence is a class of deep learning models capable of creating content that diverges from traditional discriminative models focused on interpretation or decision making." This means that generative models can produce new data like synthetic X-Rays, MRIs, and CT scans, something discriminative models can't do

Stated in the first paragraph of the introduction, of article number 1. Also, aligns with the Royal Society and Alan Turing Institute (2022) definition cited in the article, which defines synthetic data as data created by a purpose-built algorithm, confirming that data creation is the defining distinction between generative and discriminative models.

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

The article states that generative models "learn and store patterns and characteristics of the original data in their internal parameters (weights)." These trained weight act as a compressed representation of the original dataset, allowing others to generate new synthetic images without transferring actual patient data.

Stated in the Synthetic Datasets section of article number 1, in the second paragraph of the sub section Generative models, which defines "model as a dataset" as a new concept in data sharing where "sharing model weights provides an efficient alternative that allows others to generate new synthetic images with properties similar to the original data." This directly support he correct answer, as the concept replaces traditional raw data sharing with weight sharing, ruling out all other options which fall outside this definition.

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

The article states that, " Physics-informed models are primarily rule-based approaches that incorporate domain-specific knowledge and physics principles through mathematical equations and explicit constraints to generate realistic and physically plausible data." While, statistical models learn directly from data patterns and distributions, making them data-driven, not rule-based, which eliminates the option entirely.

Stated in the Synthetic Datasets section of article number 1, the article distinguishes the two model types: physics-informed models encode "expert knowledge and known physics laws" while statistical models such as VAEs, GANs, and DDPMs learn from data distributions, confirming the correct answer and ruling out all other options. In the fourth paragraph of the sub section Generative models, "Physics-informed models are primarily rule-based approaches that incorporate domain-specific knowledge and physics principles through mathematical equations and explicit constraints to generate realistic and physically plausible data. "

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

Statistical models must balance these three aspects, where no single model excels in all three simultaneously — VAEs are fast but lower quality, GANs produce high quality but risk mode collapse, and DDPMs achieve high quality and diversity but at slower speeds.

Stated in the Synthetic Datasets section of article number 1, defines the trilemma as "balancing high sample quality, comprehensive mode coverage, and rapid sampling rates." This three-way trade-off framework directly explains why different generative models are chosen for different applications, dataset generation prioritizes quality and diversity over speed, confirming that the trilemma is a practical decision-making concept, not a limitation unique to any single model type. Additionally, figure number 2 on page 4 states that, "The image generation trilemma, which represents the trade-offs between three key aspects of generative models: diversity, quality, and speed"

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

The Human Turing Test "involves domain experts who are asked to discern between real and derived medical images," providing insights into the perceptual quality and realism of generated images. Making it a human-based evaluation method, not a computational one, directly ruling out options involving processing speed, mathematical scores, anonymization, or plagiarism detection.

Stated in the Evaluating Image Quality section of article number 1, the Human Turing Test is described as the "gold standard for assessing the quality of generated medical images." Unlike computational metrics such as FID or SSIM that measure statistical properties, the Human Turing Test captures perceptual realism that mathematical scores alone cannot reflect, which is particularly critical in medical imaging where diagnostic accuracy and visual fidelity are paramount.

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

The article never makes this claim, in fact, the article warns that biases in source datasets "could be propagated or amplified in the generated data," meaning synthetic data can reduce but not permanently eliminate bias. While, all other options enhancing diversity, preserving privacy, facilitating Multi-Centre collaborations, and supporting medical education are directly mentioned as benefits in the article.

Stated in the Potentials and Promises of article number 1, Challenges and Consideration sections, the article acknowledges that synthetic data can close the fairness gap by up to 40%, but simultaneously warns that biases "could be propagated or amplified in the generated data," requiring ongoing mitigation strategies. This confirms that permanent bias elimination is never claimed making it the only option not supported by the article.

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

The article states that if a generative model "can replicate images that closely resemble the original data, the model might inadvertently reveal sensitive patient information." This is particularly concerning in medical imaging where "facial features in brain MRIs or distinctive anatomical markers in radiographs might enable reidentification even when explicit patient identifiers are removed." All other options are either technical limitations or not addressed as ethical concerns in the article.

Stated in the Challenges and Considerations of article number 1, identifies data copying as a primary ethical concern, noting that "copying happens when multiple copies of the image or captions are present in the dataset," raising serious questions about patient privacy and anonymization. And the Summary section also states that, "The challenges and ethical considerations surrounding generative artificial intelligence, including patient privacy, data copying, and potential biases that could impede clinical translation, are also addressed."

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

Stated that the FDA regulated synthetic MRI technologies "as image processing software rather than as completely novel modalities", requiring clinical validation to confirm equivalent diagnostic performance. All other options are never mentioned in the article.

Stated in the Future Directions section of article number 1, the FDA precedent establishes a validation pathway requiring "proof-of-performance equivalence, rigorous clinical validation, and postmarked surveillance", confirming that synthetic data must demonstrate diagnostic equivalence to conventional imaging before approval, serving as the key regulatory framework for future synthetic data technologies.

โจทย์ ปรนัย
What is the main purpose of the article?

States its purpose is to "highlight the similarities and differences in the epidemiology, diagnosis, and treatment of ASCVD for individuals of East Asian origin" and to identify "major knowledge gaps in our understanding of ASCVD risk." All other options like Western models, economic burden, imaging technologies, and European genetics are never addressed as the article's primary focus.

Stated in the Abbreviation and Acronyms of article number 2, "highlight the similarities and differences in the epidemiology, diagnosis, and treatment of ASCVD for individuals of East Asian origin". Stated in the abstract of article number 2, existing risk calculators consistently overestimate ASCVD risk in East Asian populations, establishing the need for region-specific risk prediction models.

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

It was developed from U.S. cohort studies and "significantly overestimate absolute CHD risk" when applied to East Asian populations. All other options were developed using native East Asian data.

Stated in the ASCVD Risk Prediction in China section of article number 2, the Framingham Risk Score was the Western baseline that East Asian countries attempted to recalibrate, but its consistent overestimation drove each country to develop their own region-specific model instead.

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

The article states that "10-year CHD event rates were 8.0% and 2.8% in Framingham men and women, respectively, compared with 1.5% and 0.6% in the CMCS men and women", demonstrating that Western cohorts had significantly higher baseline ASCVD rates, causing their models to overestimate risk when applied to East Asian populations.

Stated in the ASCVD Risk Prediction in China section of article number 2, overestimation stems from fundamental differences in "mean CHD risk and levels of major risk factors between the two cohorts", confirming that lower baseline ASCVD incidence in East Asians, not data quality, drives the overestimation.

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

Stated that China-PAR model incorporated predictors such as "geographic region and urban/rural distinctions", making it more representative of China's diverse population than Western models. All other options contradict the article, as China-PAR includes smoking as a predictor and was not derived from European trials or imaging biomarkers alone.

Stated in the ASCVD Risk Prediction in China section, Table 1 ASCVD Risk Assessment Tools in China, of article number 2, China-PAR was developed from a cohort of 27,020 individuals incorporating region-specific variables, giving it better calibration and discrimination for Chinese populations compared to the PCE, which showed "low discrimination ability and poor calibration for Chinese men."

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

Across all risks models discussed, China-PAR, Suita Score, Framingham, and Korean models, the article consistently lists age, blood pressure, cholesterol, smoking, and diabetes as core predictors. Genetic ancestry markers are never mentioned as a variable in any of the risk calculators discussed in the article.

Stated across the ASCVD Risk Prediction section of article number 2, all country specific models share common predictors including "sex, age, blood pressure, smoking, diabetes, and total cholesterol" with no model incorporating genetic ancestry markers, confirming it as the only option unsupported by the article.

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

Suita Score was selected from 10 different published risk prediction scores in Japan where "internal validation was carefully performed" using Japanese specific cohort data. The Framingham Risk Score, in contrast, was developed from U.S. community cohorts, making population origin the key distinguishing factor between the two models.

Stated in the ASCVD Risk Prediction in Japan section of article number 2, the Suita Score incorporates Japanese-specific predictors including "family history of premature CHD and impaired glucose tolerance" and notably excludes stroke. While the Framingham Risk Score which includes both CHD and stroke outcomes, confirming that population origin and outcome definition differ fundamentally between the two models.

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

The article consistently demonstrates that Western models "significantly overestimate" ASCVD risk in East Asian populations, and region-specific models using local cohort data provide better calibration and discrimination. All other options are never supported or are directly contradicted by the article.

Stated in the Future Directions and Conclusions section of article number 2, region specific models improve risk prediction by accounting for East Asia's distinct CVD profile, particularly lower CHD incidence but higher stroke rates, reducing the overestimation seen in Western models.

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

The article highlights that acculturation and environmental factors significantly shift CVD risk profiles among East Asian populations, noting differences in obesity, hypertension, and lifestyle behaviors between East Asian natives and immigrants. All other options are either contradicted or never mentioned in the article.

Stated in The impact of acculturation and environmental effects of ASCVD risk profiles section of article number 2, the article notes that "acculturation was associated with a heterogeneous pattern of CVD risk factors among Asian American subgroups" and references studies showing that East Asian immigrants report different physical activity patterns and metabolic risk profiles compared to their native counterparts, confirming that cultural and lifestyle variations, not genetic mutations or uniform healthcare access, drive regional ASCVD differences.

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

The article advocates for incorporating imaging biomarkers such as CAC scores, coronary CTA, and emerging deep learning tools alongside traditional risk factors using region-specific cohort data. All other options are directly contradicted by the article, which consistently emphasizes the need for more comprehensive and regionally validated models.

Stated in the Future Directions and Conclusions section of article number 2, the article recommends that "subclinical atherosclerosis detected by noninvasive imaging such as CAC, carotid ultrasound, and ABI should be studied more extensively in East Asian countries", highlights integrating imaging, AI, and regional data represents the proposed path forward.

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

The figure shows three distinct mechanisms: VAEs use an encoder decoder structure compressing input into latent representation, GANs use a generator discriminator system distinguishing fake from real images, and DDPMs use forward and reverse diffusion to progressively remove noise, confirming each model operates through a fundamentally different process.

Stated in the Synthetic Datasets section of article number 1 and supported by Figure 1, DDPMs "introduce noise into an image and learn to reverse this process", distinct from VAEs which compress data into latent space and GANs which use adversarial training, showing reverse diffusion as DDPMs' defining and unique mechanism.

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?

The figure shows Japan with consistently low rates in both age standardized (77) and crude CVD mortality (291 per 100,000), significantly lower than Mongolia (570/289) and North Korea (353/391). This consistency across both measures indicates that Japan's low mortality reflects genuine healthcare effectiveness, not merely population age structure differences.

Stated in the Epidemiology of ASCVD in East Asian populations living in Asia and in the United States section of article number 2, Japan's low CVD mortality is supported by the article noting that "since 1960, high stroke mortality rates but low CHD mortality rates were observed" in Japan, and that "stroke mortality has significantly improved in subsequent decades", confirming that Japan's sustained low mortality across both figure measures reflects long-term prevention efforts and healthcare system effectiveness rather than demographic factors alone.