NCA-GENM Question 97
Single answerA team is using a multimodal generative AI model to analyze medical imaging data combined with patient text records. They aim to improve the explainability of their results for doctors who rely on the model for decision-making. Which approach best utilizes multimodal models to enhance explainability in this scenario?
- A
Generate textual descriptions of the key features identified in the medical images and align them with patient text records.
- B
Use only the text data from patient records for predictions, as it is easier to explain and interpret than image data.
- C
Visualize activation heatmaps over the medical images to highlight regions of interest without integrating textual data.
- D
Combine the model's predictions with a rule-based system to ensure deterministic outputs.
Show answer and explanation
Correct answer: A
Explanation
Multimodal models are effective for improving explainability by linking different types of data, such as text and images, to provide richer and more understandable insights. In this case, generating textual descriptions from medical images and aligning them with patient records makes the predictions more interpretable for doctors.
- A. Correct.
Generating textual descriptions aligned with medical images and patient text records provides context and improves explainability by linking visual data with human-readable insights.
- B. Incorrect.
Using only text data ignores the visual information from medical images, which can be critical for accurate and explainable predictions in this scenario.
- C. Incorrect.
While activation heatmaps can highlight regions of interest in the images, they lack the additional context provided by integrating textual data.
- D. Incorrect.
Combining predictions with a rule-based system may improve consistency but does not inherently leverage the strengths of multimodal models for explainability.