This study explores the use of efficient deep learning algorithms for segmenting lower grade gliomas (LGG) in medical images. It evaluates various pre-trained atrous-convolutional architectures and U-Nets, proposing a novel transformer-based approach that surpasses traditional methods. DeepLabV3+ with MobileNetV3 backbone achieved the best results among pre-trained models, but the transformer-based approach excelled with superior segmentation accuracy and efficiency. Transfer learning significantly enhanced model performance on the LGG dataset, even with limited training samples, emphasizing the importance of selecting appropriate pre-trained models. The transformer-based method offers advantages such as efficient memory usage, better generalization, and the ability to process images of arbitrary sizes, making it suitable for clinical applications. These findings suggest that advanced deep learning techniques can improve diagnostic tools for LGG and potentially other cancers, highlighting the transformative impact of deep learning and transfer learning in medical image segmentation.

Renal ultrasound (US) imaging, also known as a kidney ultrasound, serves as a pivotal diagnostic tool for evaluating renal health. It plays a critical role in various aspects of patient care, including diagnosis, treatment planning, surgical intervention, and post-treatment monitoring, allowing healthcare providers to assess both the kidneys and bladder. Furthermore, segmentation of kidney ultrasound (US) images is essential in extracting relevant objects or areas from the complete image, aiding in the evaluation of tissue organization and improving diagnostic accuracy. While manual delineation of kidneys in US images presents challenges in clinical settings due to complexity and tedium, this study introduces a novel approach to enhance the delineation of kidney outlines in 2D ultrasound images using deep learning techniques. The method involves concatenating two DeepLabV3 models, leveraging ResNet-18 and ResNet-152 backbones. Evaluation of the method yielded promising results, with precision, recall, F1 score, accuracy, and Jaccard index reported at 0.978, 0.993, 0.985, 0.997, and 0.971 respectively, accompanied by standard deviations of ±0.009, ±0.005, ±0.005, ±0.001, and ±0.009, our findings underscore the efficacy of the proposed method in augmenting kidney outline detection in 2D ultrasound images. This advancement holds significant promise for enhancing clinical diagnosis and treatment planning in renal health management.

In recent years, Convolutional Neural Networks (CNNs) have made significant strides in the field of segmentation, particularly in semantic segmentation where both accuracy and efficiency are crucial. However, despite their high accuracy, these deep networks are not practical for real-time use due to their low inference speed. This issue has prompted researchers to explore various techniques to improve the efficiency of CNNs. One such technique is knowledge distillation, which involves transferring knowledge from a larger, cumbersome (teacher) model to a smaller, more compact (student) model. This paper proposes a simple yet efficient approach to address the issue of low inference speed in CNNs using knowledge distillation. The proposed method involves distilling knowledge from the feature maps of the teacher model to guide the learning of the student model. The approach uses a straightforward technique known as pixel-wise distillation to transfer the feature maps of the last convolution layer of the teacher model to the student model. Additionally, a pair-wise distillation technique is used to transfer pair-wise similarities of the intermediate layers. To validate the effectiveness of the proposed method, extensive experiments were conducted on the PascalVoc 2012 dataset using a state-of-the-art DeepLabV3+ segmentation network with different backbone architectures. The results showed that the proposed method achieved a balanced mean Intersection over Union (mIoU) and training time.

Semantic segmentation is a fundamental task in computer vision, requiring precise object delineation for applications such as autonomous driving and medical imaging. Traditional convolutional neural networks (CNNs) often struggle with capturing long-range dependencies and preserving fine spatial details. It is the study's goal to make segmentation more accurate by adding adaptive attention to the encoder and decoder stages of the U-Net-based architecture. The proposed network employs ResNet-50 as its backbone for efficient multi-level feature extraction. The encoder incorporates an Efficient Channel Attention Atrous Spatial Pyramid Pooling (ECA-ASPP) module to enhance its context representation. This module uses dilated convolutions and adaptive channel attention to improve the collection of features at different sizes. There is also a Point-wise Spatial Attention (PSA) module in the decoder that dynamically gathers global contextual information while keeping fine-grained spatial details. Extensive experiments on the Stanford Background Dataset demonstrate a consistent improvement across all segmentation categories. The best-performing model achieves a mean Intersection over Union (mIoU) of 78.65%, outperforming baseline approaches. Furthermore, evaluation on the Cityscapes dataset yields an mIoU of 80.46%, surpassing state-of-the-art methods such as DeepLabV3 and DANet. These results show that using adaptive attention during both the encoding and decoding steps works well, finding a good balance between accurate segmentation and fast computing. The proposed network demonstrates strong potential for real-world applications requiring precise segmentation.

Purpose: Glioma tumor segmentation is an essential step in clinical decision making. Recently, computer-aided methods have been widely used for rapid and accurate delineation of the tumor regions. Methods based on image feature extraction can be used as fast methods, while segmentation based on the physiology and pharmacokinetic of the tissues is more accurate. This study aims to compare the performance of tumor segmentation based on these two different methods. Materials and Methods: Nested Model Selection (NMS) based on Extended-Toft’ s model was applied to 190 Dynamic Contrast-Enhanced MRI (DCE-MRI) slices acquired from 25 Glioblastoma Multiforme (GBM) patients in 70 time-points. A model with three pharmacokinetic parameters, Model 3, is usually assigned to tumor voxel based on the time-contrast concentration signal. We utilized Deep-Net as a CNN network, based on DeepLabV3+ and layers of pre-trained resnet18, which has been trained with 17288 T1-Contrast MRI slices with HGG brain tumor to predict the tumor region in our 190 DCE MRI T1 images. The NMS-based physiological tumor segmentation was considered as a reference to compare the results of tumor segmentation by Deep-Net. Dice, Jaccard, and overlay similarity coefficients were used to evaluate the tumor segmentation accuracy and reliability of the Deep tumor segmentation method. Results: The results showed a relatively high similarity coefficient (Dice coefficient: 0. 73± 0. 15, Jaccard coefficient: 0. 66± 0. 17, and overlay coefficient: 0. 71± 0. 15) between deep learning tumor segmentation and the tumor region identified by the NMS method. The results indicate that the deep learning methods may be used as accurate and robust tumor segmentation. Conclusion: Deep learning-based segmentation can play a significant role to increase the segmentation accuracy in clinical application, if their training process is completely automatic and independent from human error.