AI-Powered Segmentation of Tumors in Multi-Modal Medical Images

Precise tumor segmentation in medical images is essential for the efficient diagnosis and treatment planning process. The current work presents a novel AI-based framework that allows for automated tumor segmentation from multi-modal images comprising MRI, CT, and PET scans. To enhance segmentation accuracy, a hybrid deep learning model integrating the capabilities of U-Net in extracting spatial features and DenseNet for contextual learning is proposed. The framework exploits the complementary information from the inputs and provides precise and robust results.The model was tested with a curated dataset, where its Dice Similarity Coefficient is <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 2. 8 \%}$</tex> and its IoU score is <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 7. 3 \%}$</tex>, outperforming methods available in the literature. Furthermore, the system has post-segmentation tumor volume estimation, with malignancy classification at a level of 94% in malignancy prediction.