A recent study from multiple prestigious institutions concludes that AI is not yet capable of making clinical diagnoses from radiological scans, despite the belief in its potential within radiology. Researchers identified shortcomings in existing datasets, particularly their size and variety, which limit training capabilities. To counter this, they created the DeepTumorVQA benchmark — consisting of 9,262 CT volumes to assess AI’s performance in understanding abdominal tumors. This benchmark aims to evaluate visual language models more effectively with categories like Recognition and Visual Reasoning.
AI models are not yet capable of making clinical diagnoses from radiological scans, as demonstrated by a collaborative study from several leading universities.
The benchmark test DeepTumorVQA was developed to better assess AI's ability to interpret abdominal tumor CT scans with over 9,000 volumes and expert-crafted questions.
Collection
[
|
...
]