SPEEDNet: Salient Pyramidal Enhancement Encoder-Decoder Network for Colonoscopy Images

Accurate identification and precise delineation of regions of significance, such as tumors or lesions, is a pivotal goal in medical imaging analysis. This paper proposes SPEEDNet, a novel architecture for precisely segmenting lesions within colonoscopy images. SPEEDNet uses a novel block named “Dilated-Involutional Pyramidal Convolution Fusion” (DIPC). A DIPC block combines the dilated involution layers pairwise into a pyramidal structure to convert the feature maps into a compact space. This design lowers the total number of parameters while improving the learning of representations across an optimal receptive field. The proposed DIPC Block prioritizes the integration of salient maps and feature maps from the preceding stage of the encoder through a direct feed-forward connection. We have employed Tversky Loss to handle the class imbalance challenge and manage the trade-off between precision and recall. On the EBHISeg dataset, SPEEDNet outperforms five state-of-art networks, viz., UNet, FeedNet, SegNet, MedT and AttesResDUNet. Precisely, SPEEDNet attains an average dice score of 0.952 and a recall of 0.971. Qualitative results and ablation studies provide additional insights into the effectiveness of SPEEDNet. The model size of SPEEDNet is 9.8 MB, significantly smaller than that of UNet (119.8 MB), FeedNet (61.8 MB), and AttesResDUNet (139.1 MB). SPEEDNet has a throughput of 33 frames per second on the GeForce MX330 GPU.