Rao, P.K.; Chatterjee, S.; Janardhan, M.; Nagaraju, K.; Khan, S.B.; Almusharraf, A.; Alharbe, A.I. Optimizing Inference Distribution for Efficient Kidney Tumor Segmentation Using a UNet-PWP Deep-Learning Model with XAI on CT Scan Images. Diagnostics 2023, 13, 3244. https://doi.org/10.3390/diagnostics13203244
Rao, P.K.; Chatterjee, S.; Janardhan, M.; Nagaraju, K.; Khan, S.B.; Almusharraf, A.; Alharbe, A.I. Optimizing Inference Distribution for Efficient Kidney Tumor Segmentation Using a UNet-PWP Deep-Learning Model with XAI on CT Scan Images. Diagnostics 2023, 13, 3244. https://doi.org/10.3390/diagnostics13203244
Rao, P.K.; Chatterjee, S.; Janardhan, M.; Nagaraju, K.; Khan, S.B.; Almusharraf, A.; Alharbe, A.I. Optimizing Inference Distribution for Efficient Kidney Tumor Segmentation Using a UNet-PWP Deep-Learning Model with XAI on CT Scan Images. Diagnostics 2023, 13, 3244. https://doi.org/10.3390/diagnostics13203244
Rao, P.K.; Chatterjee, S.; Janardhan, M.; Nagaraju, K.; Khan, S.B.; Almusharraf, A.; Alharbe, A.I. Optimizing Inference Distribution for Efficient Kidney Tumor Segmentation Using a UNet-PWP Deep-Learning Model with XAI on CT Scan Images. Diagnostics 2023, 13, 3244. https://doi.org/10.3390/diagnostics13203244
Abstract
Motivation: It is essential for the diagnosis and treatment of renal cancers to segment kidney tumours precisely and effectively. In medical image segmentation tasks, deep learning models have demonstrated promising results, and the UNet model is widely employed in this field. However, optimising the UNet model for kidney tumour segmentation further can improve its efficacy and deployment feasibility. Related Works: Previous works have explored various techniques to improve the efficiency of deep learning models in medical image segmentation. Image partitioning methods divides the input image into smaller regions, enabling parallel processing and reducing memory requirements. Pruning techniques eliminates redundant or insignificant weights, neurons and connections, resulting in a more compact and efficient model architecture. However, the UNet model architecture and its complexity in the context of kidney tumor segmentation using UNet remains unexplored. Methodology: The proposed methodology consists of adaptive partitioning and weight pruning. Adaptive partitioning divides the UNet model into smaller sub-models, facilitating parallel processing and accelerating inference without compromising segmentation accuracy. Weight pruning techniques remove redundant or less significant weights from the UNet-P model, reducing complexity and improving inference time by eliminating unnecessary computations. The adaptive partitioning and weight pruning processes are seamlessly integrated within the UNet-P architecture, resulting in an optimized model for kidney tumor segmentation. Results: We performed experiments utilising a dataset of KiT 23 CT scan images containing images of kidney malignancies. Compared to the standard UNet model, the optimised UNet-PWP model with adaptive partitioning and weight pruning obtained significant efficiency gains. The adaptive partitioning permitted parallel computation of sub-models, which accelerated inference times. Weight pruning decreased the UNet-PWP model’s complexity without compromising segmentation accuracy, thereby enhancing efficiency. The results of our experiments demonstrated the efficacy of the proposed method with 98% accuracy, demonstrating its potential for deployment in health sector.
Computer Science and Mathematics, Artificial Intelligence and Machine Learning
Copyright:
This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
,
*
,
,
