International Journal of Drug Delivery Technology · Jun 8, 2026
Edge-Deployable AI for Lung Cancer Screening: A Lightweight Deep Learning Approach
By Kurra M, Sharma O, Ghantasala GSP
The performance of deep learning models for early detection of lung cancer screening in clinical settings with limited resources is still hindered due to the heavy computational requirements associated with deep neural networks. To alleviate this problem, we propose a compact model optimization technique for effective deep learning-based real-time CT image-based lung cancer detection. In this work, we utilize INT8 quantization (FP32 to INT8 with TensorRT) in combination with knowledge distillation, where a pretrained ResNet-34 model acts as the teacher, providing guidance to a compact MobileNetV3 network so that it can maintain discriminative power while keeping computational efficiency high. The proposed model is evaluated on the NVIDIA Jetson AGX Xavier platform and outperforms the existing lightweight models in terms of both model size (42%) and computations (37%), surpassing lightweight architectures like LBC-CNN, EA-LungNet, and MSWindDYOLO. Experiments conducted using the external dataset IQ-OTHNCCD reveal that the accuracy drop of the optimized model in comparison with the initial one is just 0.7%, and at the same time, its speed is quite good (on average, 86 ms per test scan), exceeding the 100 ms threshold.