Optimizing Inference Distribution for Efficient Kidney Tumor Segmentation using a UNet-PWP Deep Learning Model on CT Scan Images

• DOI: 10.20944/preprints202308.0888.v1
• Publication Date: 2023-08-14T00:31:10Z
• Authors: Patike Kiran Rao; Dr Subarna Chatte...; Dr M Janardhan; Dr Nagaraju K; Dr. Surbhi Bhatia...; Dr. Ahlam Almusha...; Abdullah I Alharbe
• Keywords: Pruning

Motivation: It is essential for the diagnosis and treatment of renal cancers to segment kidney tumours precisely effectively. In medical image segmentation tasks, deep learning models have demonstrated promising results, UNet model widely employed in this field. However, optimising tumour further can improve its efficacy deployment feasibility. Related Works: Previous works explored various techniques efficiency segmentation. Image partitioning methods divides input into smaller regions, enabling parallel processing reducing memory requirements. Pruning eliminates redundant or insignificant weights, neurons connections, resulting a more compact efficient architecture. architecture complexity context tumor using remains unexplored. Methodology: The proposed methodology consists adaptive weight pruning. Adaptive sub-models, facilitating accelerating inference without compromising accuracy. Weight pruning remove less significant weights from UNet-P model, improving time by eliminating unnecessary computations. processes are seamlessly integrated within architecture, an optimized Results: We performed experiments utilising dataset KiT 23 CT scan images containing malignancies. Compared standard optimised UNet-PWP with obtained gains. permitted computation which accelerated times. decreased model’s accuracy, thereby enhancing efficiency. results our method 98% demonstrating potential health sector.