Lightweight Deep Learning Models for Kidney Tumor Segmentation in Resource
Rajagopal K, N.Muthuvairavan Pillai, V Sheeja Kumari, A. Rekha, Sriram S, R. Vijayalakshmi · 2025
Early detection and treatment planning depend heavily on kidney tumour segmentation; yet, deep learning models are generally computationally intensive and therefore unsuitable for situations with limited resources. Developing and accessing lightweight deep learning models tailored for kidney tumour segmentation with high accuracy and minimal processing overhead is the goal of this project. Creating effective architectures, reducing model size and inference time, and preserving segmentation accuracy on par with state-of-the-art models are among the goals. To improve performance while preserving efficiency, the suggested method makes use of depthwise separable convolutions, attention mechanisms, and knowledge distillation. Preprocessing methods were employed to increase segmentation accuracy in a dataset of kidney tumour pictures used for training and validation. The models were assessed using measures for computational efficiency, intersection over union (IoU), and dice similarity coefficient (DSC). According to the results, the optimised models can be used in clinical situations with limited resources because they provide competitive segmentation accuracy while drastically lowering memory footprint and inference time. This paper offers a practical answer for resource-constrained applications by proving the viability of lightweight deep learning models in medical image segmentation.