Compositional Segmentation of Cardiac Images Leveraging Metadata

Our Contributions

• We propose a novel compositional segmentation approach that simultaneously localizes the heart (super segmentation) and segments the heart structures (sub-segmentation).
• We propose a Cross-Modal Feature Integration (CMFI) module to utilize the image metadata, including acquisition parameters, medical condition, and demographic of the patient to conditionally modulate the segmentation network.
• Extensive quantitative and qualitative experimental comparisons demonstrate that our proposed method outperforms the existing state-of-the-art. We evaluate the proposed approaches on two different modalities, MRI and ultrasound, and show that our approach excels in these diverse domains. The consistent performance improvements observed in both modalities indicate that our method could yield similar accu

Paper Available Here