Work

Enhanced ResNet-50 with restraint-defined feature learning for automated lung cancer classification from MRI images

Year: 2026
Type: article
Abstract: Despite the progress of medical image analysis, increasingly accurate classification of lung cancer is still a difficult task, especially due to the variability of the appearance of images, the tissue properties and the discriminative patterns of different classes. In this study, an improved deep learning framework is presented, RDFV-EResNet50, which includes a ResNet-50 backbone, three extra residual blocks, squeeze-and-excitation (SE) attention, and Restraint-Defined Feature Vectors (RDFVs) for lung image classification in magnetic resonance imaging (MRI). The residual architecture is designed to enhance hierarchical feature learning, and SE mechanism adaptively recalibrates channel-wise feature responses. The feature weight using the RDFV method is embedded to highlight discriminative representations prior to classification. The framework is tested by using the performance values reported in the experimental study against MMDLF-LCS-HI, InceptionNeXt and YOLOv8-TNM. The manuscript gives an accuracy of 99.3%, precision of 98%, recall of 97%, and F1 score of 97.5% for the proposed approach. The results reported herein must be viewed in light of the composition of the data set, the partitioning protocol, and the evaluation procedure. Additional controlled ablation studies, confusion-matrix analyses of the models, and external validation are required to determine the contribution of each component and the generalizability of the model. The proposed framework would serve as a framework for further investigation of deep learning based lung image classification methods, yet there needs to be further validation before it is applicable in a clinical setting. Clinical trial number : Not applicable.
Language: English
Open Access status: gold
Estimated APC paid: $1,520
OpenAlex ID: W7235349134
FWCI:
Cites: 29
Cited by:
Related to: 10