Modeling an effectual multi-section You Only Look Once for enhancing lung cancer prediction

The shape and size of nodules are significant pointers of malignancy in cancer diagnosis. Moreover, considerable capture of nodules' structural data attained from computed tomography (CT) scans in computer-aided design is a confronting task. Various investigations deal with computationally deep ensemble approaches/convolutional neural network (CNN) models; however, sampling tumors based on multi-section-You Only Look Once (MS-YOLO) architecture, this anticipated model acquires nodules' multi-sections from various views and encodes nodule's information to data aggregation from diverse cross-sections through pooling layers. Subsequently, the features are utilized for the nodule classification task. This MS YOLO does not need any nodules' spatial annotation. However, it works directly over the cross-section acquired from enclosed nodule volume. This work has been analyzed using LUNA R16 dataset. It attains adequate performance with a mean value of 90.8% classification accuracy. Anticipated architecture is cast-off to choose cross-section determination of malignancy that helps in output interpretation. The proposed model shows better trade-off among previous methodologies. Simulation was carried out in MATLAB environment.

Automated summary

The study proposes a new nodule diagnosis model, MS YOLO, which achieves high accuracy classification without the need for spatial annotation of nodules. By sampling multiple cross-sections of the tumor and using convolutional neural networks for processing, the research demonstrates better performance and achieves high classification accuracy in experiments.