High-Efficiency Video Coding provides a better compression ratio compared to earlier standard, H.264/Advanced Video Coding. In fact, HEVC saves 50 bit rate compared to H.264/AVC for the same subjective quality. This improvement is notably obtained through the hierarchical quadtree structured Coding Unit. However, the computational complexity significantly increases due to the full search Rate-Distortion Optimization, which allows reaching the optimal Coding Tree Unit partition. Despite the many speedup algorithms developed in the literature, the HEVC encoding complexity still remains a crucial problem in video coding field. Towards this goal, we propose in this paper a deep learning model-based fast mode decision algorithm for HEVC intermode. Firstly, we provide a deep insight overview of the proposed CNN-LSTM, which plays a kernel and pivotal role in this contribution, thus predicting the CU splitting and reducing the HEVC encoding complexity. Secondly, a large training and inference dataset for HEVC intercoding was investigated to train and test the proposed deep framework. Based on this framework, the temporal correlation of the CU partition for each video frame is solved by the LSTM network. Numerical results prove that the proposed CNN-LSTM scheme reduces the encoding complexity by 58.60 with an increase in the BD rate of 1.78 and a decrease in the BD-PSNR of -0.053dB. Compared to the related works, the proposed scheme has achieved a best compromise between RD performance and complexity reduction, as proven by experimental results.

Despite advancements in imaging technologies, traditional diagnostic methods relying on manual interpretation by clinicians remain prone to errors. This underscores the need for robust automated segmentation techniques. Accurate three-dimensional (3D) Magnetic Resonance Imaging (MRI) brain tumor segmentation is critical for effective diagnosis in neuroimaging. Given the complexity of brain tumors, deep learning approaches have emerged as more suitable than traditional methods for efficient and precise segmentation. The dual difficulties of high computing needs in 3D convolutional networks and the necessity of more precise brain tumor diagnosis are addressed by this work. We provide an artificial intelligence (AI) framework combining bi-directional convolutional long short-term memory (ConvLSTM) layers with densely connected convolutions (DCC). Our method enhances spatial feature learning employing dense connections, and catches complex temporal links across MRI slices. Particularly for the identification of the Enhancing Tumor (ET) region, this novel combination solves the complexity limitation of convolutional techniques, hence improving segmentation robustness. Our approach is validated by extensive testing on the Brain Tumor Segmentation (BraTS) 2020 and 2021 datasets. On the BraTS 2021 dataset, the proposed M-BDCU-Net model beats many state-of-the-art approaches for particular tumor sub-regions with Dice Similarity Coefficient (DSC) scores of 0.81 for Tumor Core (TC), 0.85 for Enhancing Tumor (ET), and 0.82 for Whole Tumor (WT). With 8.81 million parameters and 60.34 GFLOPs, our model also greatly lowers complexity by up to 15% and balances accuracy with efficiency. It is thus much lighter than current models. These results highlight how rapid, more dependable, and exact brain tumor detection our system could allow. This work fills important voids in present research by combining methodological innovation with exhaustive evaluation and provides a useful answer for therapeutic uses.