| 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 |
|
Some AI models have been trained on language and various medical images, such as in the field of radiology and ophthalmology, which allows them to analyze these images and aid doctors in making proper diagnoses.
|
The text states that some AI models, such as Med-Gemma and MedImageInsight have been trained on medical images such as radiology, dermatology, and ophthalmology images, and are called multimodal foundational models.
|
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 |
|
These models interpret patterns and classify them from already existing data, however, they also create new data or images with similarities to the original data. They then use these instead to increase efficiency and generate new images that are have characteristics of the old data.
|
Generative models store patterns and other features of the data, which create weights that contain the features and relationships of the used data. This then allows it to generate new images with similarities to the old data.
|
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 |
|
These models analyze the key characteristics and the trends of the data, and then store these as weights, which allow them to replicate this data in their own way, and create new data. This new data helps with applying the results while still being accurate because it follows the relationships of the old data.
|
Traditional dataset sharing transfers actual data and real images, but these model weights create artificial data that still maintains the key characteristics of the data that was analyzed. These datasets have similar correlation and features to the original data.
|
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 |
|
These use already existing knowledge and theories within fields to simulate scenarios instead of using real data. This helps in applications rather than having to interpret raw data. Statistical models, however, learn from actual data and statistics, following these patterns and relationships shown.
|
Physics models use mathematical equations and physics to simulate what would actually happen in different situations. They use knowledge from experts in the field instead of data as a simulation. This can be useful in many scenarios, such as blood flow, and radiotherapy.
|
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 |
|
Different types of models based on statistics can have different applications and benefits. Some have quick sampling abilities, but the speed sometimes gives worse sample quality. Others have high quality samples, but cannot sample many things that once. Some excel in generating samples through a wide range, but are slower than the others. These all have a tradeoff and are better at different things.
|
The image of the image generation trilemma is a triangle between quality, speed, and diversity, and includes models with different advantages in each part. VAEs are fast and diverse, but aren't that high quality. GANs are fast and have good quality, but don't have good diversity. DDPMs have good quality and diversity, but are slow.
|
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 |
|
Experts are used to verify these results and see if they are actually possible and accurate. This allows for proper checking of accuracy.
|
Domain experts are brought in to check between real images and generated images to see if they are accurate or comparable to the original one.
|
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?
|
Supporting medical education |
|
The article states that "researchers can unlock unprecedented levels of data diversity, privacy preservation, and multifuncitonality," which does not include anything about medical education.
|
This is found under the section "Potentials and promises".
|
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 |
|
Models trained on specific datasets can generate images that are too similar to the actual conditions, which is classified and private information.
|
The models can reveal patient information if they generate images too similar to the ones from the data they were trained on. This is found in section "Patient privacy and data copying".
|
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 |
|
The FDA regulated MRI technologies and had a lot of evaluation in making sure that radiologists still had accurate results when using images generated by AI.
|
This came from the FDA clearance of synthetic MRI technologies.
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 10 |
What is the main purpose of the article?
|
To introduce new diagnostic imaging technologies |
|
Models that generate images based on existing data sets have the potential to impact the research in terms of imaging and can be very helpful if done right.
|
This comes from the conclusions that state that overcoming challenges associated with AI will allow humans to use the full potential of AI in medical imaging research and in improving healthcare.
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|
| 11 |
Which of the following models was originally developed for a Western population?
|
Korean Risk Prediction Model (KRPM) |
|
The data was analyzed from Koreans living in the U.S. at first.
|
This data came from the Koreans in America.
|
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 |
|
The current models do not account for race and East Asians are underrepresented or in lower populations, so it cannot be generally applied to them.
|
The model made from the AHA PREVENT does not factor in race.
|
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 found that the Western model overestimated the rate in Chinese people.
|
It used data from China.
|
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?
|
Age |
|
They factored in everything except age.
|
This came from the Central Illustration: East Asian Cardiovascular Risk Calculators.
|
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 |
|
It came from Japanese data.
|
The Framingham Score uses Western metrics.
|
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 |
|
Western models overestimate the risk in East Asians.
|
Developing new models for Asia will allow the proper risk analysis to be made for that specific country.
|
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 |
|
The lifestyle of Asians is different from other races in Europe and the U.S..
|
This leads to underrepresentation and Asians are also a small part of the population in the Western countries.
|
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 |
|
Regional data makes sure that the results are accurate for that specific group.
|
This will help make sure that the results are accurate and factors in the lifestyles and proper metrics for each race.
|
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. |
|
It uses a reverse process to decode images.
|
The image shows the arrows pointing forward and then backward which means reversing the process to get the image.
|
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?
|
Mongolia and North Korea demonstrate higher CVD mortality due to older population structures alone. |
|
This is the only one that is accurate within the graph, but no data is given in terms of the actual ages of the population or the size of the population.
|
The other answers do not fit the data shown.
|
7 |
-.50
-.25
+.25
เต็ม
0
-35%
+30%
+35%
|