A Real-Time Lung Tumor Detection Approach Using YOLOv11 and Vision Transformers

Computer-aided diagnostic systems have proven efficient in accurately and effectively diagnosing lung cancer. How- ever, the speed of diagnosis remains a prime challenge for spe- cialists who need immediate results. This study proposes a novel approach for rapidly detecting lung cancer through CT scan images. The proposed approach relies on YOLOv11 for region-of- interest detection integrated with Vision Transformer for feature extraction and classification. Three distinct datasets were used for training, validation, and testing. Thus, an unseen dataset was used to further assess the model's performance. Moreover, Local Interpretable Model-Agnostic Explanations (LIME) was used to enhance the model's trustworthiness and comprehensibility. The proposed method achieved high performance, completing training and validation in about half a minute, with an accuracy of 92% and a precision of 96%. When tested on the DLCT dataset, it achieved a performance of 97.86%. The proposed model achieved very promising results with real-time diagnosis speed.