A study presents a new lightweight deep learning model that can classify lung cancer from CT scans with very high accuracy while using far less computing power than traditional methods.
Researchers combined EfficientNetB0 with a lightweight attention system and a capsule-style feature module to better detect and analyze lung nodules. Unlike older capsule networks, the model removes complex “dynamic routing,” making it faster and easier to run.
On the IQ-OTH/NCCD dataset, the system achieved 100% accuracy in distinguishing benign, malignant, and normal cases, while also cutting computation by about 96% and reducing inference time from 3 seconds to 0.124 seconds. The model is small enough to run on standard computers and low-cost GPUs, making it useful for hospitals with limited resources.
However, performance dropped when tested on other datasets due to differences in CT scan methods between hospitals. Researchers now aim to improve its ability to generalize across different clinical settings.
Overall, the tool could help radiologists detect lung cancer more quickly and efficiently in real-world practice.