Search

To establish a multivariate linear regression model for predicting the difficulty of high-intensity focused ultrasound (HIFU) ablation of uterine fibroids based on multi-sequence magnetic resonance imaging radiomics features. A retrospective analysis was conducted on 218 patients with uterine fibroids who underwent HIFU treatment, including 178 cases from Yongchuan Hospital of Chongqing Medical University and 40 cases from the Second Affiliated Hospital of Chongqing Medical University (external validation set). Radiomics features were extracted and selected from magnetic resonance images, and potentially related imaging features were collected. The energy efficiency factor (EEF) was used as the dependent variable. Imaging models, radiomics models, and joint models were established using a stepwise approach. The model with the highest R2 value was selected for external validation. The R2 value of the combined model was 0.642, higher than that of other models. Spearman correlation analysis showed a correlation coefficient of R = 0.824 (P < 0.001) between predicted EEF and actual EEF. External validation yielded a correlation coefficient of R = 0.645 (P < 0.001). A model for predicting EEF has been developed, which is clinically important for predicting the difficulty of HIFU treatment of uterine fibroids. [ABSTRACT FROM AUTHOR]

In clinical diagnosis, magnetic resonance imaging (MRI) plays a vital role in analyzing soft tissues. However, the images are affected by noise that is random in nature. The noise affects the quality of the image, which impacts the accuracy of diagnosis. To address this, in this paper, four enhanced versions of filters: enhanced Lee filter (ELF), enhanced Frost filter (EFF), enhanced Kuan filter (EKF) and enhanced Boxcar filter (EBCF) are used. The enhanced image features are used for classification with three different CNN classification models: Convolutional Neural Network (CNN), CNN with attention module (CNN-AM) and CNN with a residual module (CNN-RM). The enhanced filters help in improving the features. In the proposed architecture, the spatial attention module achieves these benefits by applying operations such as average pooling, max pooling, and sigmoid activation to selectively highlight key spatial regions in MRI images, leading to improved classification accuracy and overall performance. The proposed methods are compared with the state-of-the-art methods, and they ensure better image enhancement with high PSNR values of 32.1, 36.7 and 39.1, respectively. The classification models ensure high accuracy of 94%, 95% and 98%.