Automatic Recognition Algorithm for Renal Tumors and Cysts in CT Images using Mamba and YOLO11.

Abstract

Introduction: Renal tumors pose a serious threat to patient health and survival, highlighting the importance of early detection and accurate diagnosis. In clinical practice, differentiating renal tumors from cysts in CT images remains challenging due to similar imaging characteristics and complex anatomical structures. The aim of this study is to develop an improved detection method for renal tumors and cysts based on an enhanced YOLO11 framework.

Methods: An improved YOLO11-based detection model incorporating a Mamba-inspired architecture is proposed. A gated state-space modeling module is introduced into the backbone network to enhance the modeling capability of spatial and channel information and effectively focus on key regional features. A Dynamic Upsampling module (DySample) is then adopted in the neck network to improve multi-scale feature fusion. In addition, a Multi-Dimensional Local Channel Attention (MLCA) mechanism is integrated before the detection head to jointly refine spatial and channel features, thereby enhancing the localization capability for lesion areas.

Results: Experimental results demonstrate that the proposed method achieves a precision of 0.837, a recall of 0.636, mAP@0.5 of 0.732, and mAP@0.5:0.95 of 0.505. Compared with the YOLO11 model, these metrics are improved by 3.1%, 0.1%, 2.1%, and 2.5%, respectively, indicating an overall enhancement in detection performance.

Discussion: YOLO11-Mamba has achieved improvements in detection accuracy and localization performance, but there are still some potential limitations. Among these, the introduction of state space models and attention mechanisms has increased the model's parameter count and computational complexity to some extent, which may pose challenges for clinical deployment, pointing the way for future research.

Conclusion: The proposed method demonstrates effective performance in the detection of renal tumors and cysts from CT images. The results show fewer missed detections and improved lesion localization accuracy, suggesting the proposed model is a promising tool for renal lesion detection and clinical imaging.