TBConvL-Net: A hybrid deep learning architecture for robust medical image segmentation

Shahzaib Iqbal, Tariq Mahmood Khan, Syed Saud Naqvi, Asim Naveed, Erik Meijering

Abstract: Highlights•Design of a hybrid network that takes into account the local features of a CNN.•BConvLSTM and the Swin Transformer capture temporal and channel dependencies.•Composite loss assesses segmentation robustness and boundary agreement for evaluation.•Depth-wise separable convolutions reduce computational load and enhance feature learning, while the optimal number of filters prevents filter overlap and promotes convergence.•The proposed method is evaluated for 7 segmentation applications using 10 public datasets.