Alapfogalmak
Developing a groundbreaking medical multimodal-multitask foundation model (M3FM) revolutionizes chest CT imaging performance, enabling superior outcomes in lung cancer screening and other related tasks.
Kivonat
The article introduces the M3FM model, emphasizing its ability to synergize multimodal data for diverse clinical tasks, particularly in chest CT imaging. By curating a comprehensive dataset of 163,725 3D chest CT exams and applying a multimodal question-answering framework, M3FM outperforms single-modality models. The model can identify informative data elements relevant to specific clinical tasks and adapt to new tasks with small datasets. It handles different combinations of incomplete multimodal datasets and high-dimensional medical images effectively. The study highlights the importance of AI models in healthcare and aims to improve precision medicine through advanced technology.
Statisztikák
NLST participants: 26,254 patients with LDCT scans.
MIDRC dataset: 35,730 volumetric chest CT series from 7,609 patients.
MGH dataset: 905 patients with chest CT series.
OpenM3Chest dataset includes 17 training, 17 validation, and 33 testing datasets for various medical tasks.
Idézetek
"Our M3FM model significantly improves the performance of the counterpart model trained on single-modality data for individual tasks."
"M3FM can be adapted to boost the performance of new tasks with a small out-of-distribution dataset."
"Our extensive experimental results demonstrate that by synergizing multimodal data and multitasks, our M3FM model significantly improves the performance of the counterpart model trained on single-modality data for individual tasks."