For the deployment of Sixth Generation (6G) networks, integrating Massive Multiple-Input Multiple-Output (Massive MIMO) systems with Intelligent Reflecting Surfaces (IRS) is highly recommended due to its significant benefits in reducing communication losses for Non-Line-of-Sight (NLoS) conditions. However, the use of passive IRS presents challenges in channel estimation, mainly due to the significant feedback overhead required in Frequency Division Duplex (FDD)-based Massive MIMO systems. To address these challenges, this paper introduces a novel Denoising Gated Recurrent Unit with a Dropout-based Channel state information Network (DGD-CNet). The proposed DGD-CNet model is specifically designed for FDD-based IRS-aided Massive MIMO systems, aiming to reduce the feedback overhead while improving the channel estimation accuracy. By leveraging the Dropout (DO) technique with the Gated Recurrent Unit (GRU), the DGD-CNet model enhances the channel estimation accuracy and effectively captures both spatial structures and time correlation in time-varying channels. The results show that the proposed DGD-CNet model outperformed existing models in the literature, achieving at least a 26% improvement in Normalized Mean Square Error (NMSE), a 2% increase in correlation coefficient, and a 4% in system accuracy under Low-Compression Ratio (Low-CR) in indoor situations. Additionally, the proposed model demonstrates effectiveness across different CRs and in outdoor scenarios.

Intelligent reflecting surfaces (IRS) are an innovative technique that dramatically increases system efficiency. The integration of massive multiple-input multiple-output (massive MIMO) and IRS has been considered the most efficient route to 6G networks. An important challenge in IRS-aided massive MIMO wireless systems is channel estimation. With a rise in the number of IRS-reflecting elements and IRS-assisted users, channel training overhead becomes too large, resulting in large transmission delays and poor data transfer rates. To overcome this problem, an enhanced compressive sensing (CS) method to determine reliable channel state information (CSI) in IRS-aided massive MIMO systems is proposed, which combines enhanced compressive sensing with a deep denoising convolution neural network (CsiNet-DeCNN). By using deep learning methods to denoise channel data, our proposed model is validated numerically, indicating that it is accurate with low NMSE. Further, the results indicate that CsiNet-DeCNN performs better than traditional CS methods in estimating channel parameters.

MRI-guided radiotherapy (MRIgRT) enables the real-time visualization of tumor motion, allowing adaptive radiation delivery based on dynamic anatomical changes. However, respiratory-induced tumor motion remains a major challenge, particularly for thoracic and abdominal tumors. Continuous tumor motion may reduce treatment accuracy and increase radiation exposure to surrounding healthy tissues. Therefore, reliable and efficient tumor tracking is essential for real-time motion management in MRI-guided radiotherapy. Recent advances in artificial intelligence have demonstrated significant potential for medical image analysis; however, many existing tumor tracking approaches rely on segmentation-based methods that require detailed annotations and complex processing, which can limit their use in real-time clinical environments. In this work, we propose a detection-based tumor tracking framework that integrates the YOLOv11 object detection model with an enhanced StrongSORT tracking algorithm (EA-StrongSORT). The proposed approach replaces the conventional re-identification backbone with a lightweight EfficientNetV2 architecture augmented with an Efficient Channel Attention (ECA) mechanism. The overall framework follows a tracking-by-detection concept, where tumor regions are first detected and subsequently associated across frames. The proposed framework is evaluated on the TrackRAD2025 dataset using multiple YOLOv11 variants to analyze the balance between performance and model complexity. Experimental results demonstrate that the lightweight YOLOv11n model achieves the best detection performance, with a precision of 0.912, recall of 0.607, mean Average Precision (mAP) of 0.771, and ${\mathrm{mAP}}_{50–95}$ of 0.608. Furthermore, the proposed tracking framework achieves stable temporal association, with Multiple-Object Tracking Accuracy (MOTA) scores exceeding 91% and Higher-Order Tracking Accuracy (HOTA) scores around 90%. The proposed framework demonstrates the potential of detection-based tumor localization and tracking for real-time MRI-guided radiotherapy applications.