Authors: Peng, Quchen | Yan, Yunqi | Qian, Lijun | Suo, Shiteng | Guo, Yi | Xu, Jianrong | Wang, Yuanyuan
Article Type: Research Article
Abstract: BACKGROUND: The incidence of liver tumors is among the top three in China. The treatments of benign and malignant tumors are different. Accurate diagnosis plays an important role in guiding the treatment of tumors. OBJECTIVE: The aim of this study is to solve the following: (1) blurred boundary between the liver tumor and other organs causes incorrect segmentation of liver tumor boundaries; (2) large difference in tumor size and the diversity in texture and grayscale are major challenges in liver tumor classification tasks. METHODS: Firstly, the liver tumor is segmented from the original CT images by a tumor segmentation network, …UNet++ with fusion loss and atrous spatial pyramid pooling (FLAS-UNet++). The proposed segmentation method can solve the problem of tumor edge segmentation error by learning the tumor edge information. Secondly they are adaptively cropped according to the tumor volume to reduce the over-fitting and over-sensitivity of the deep network. Thirdly an improved Dense Block is designed to pay more attention to the changes in grayscale and texture between benign and malignant tumors. Finally, the features extracted from the network combined with tumor volume, patient’s sex and age, are sent to a classifier for diagnosis. RESULT: Liver tumor segmentation results show that the dice, HD95 reached 71.9%, 12.1 mm, respectively. The classification results show that the accuracy, specificity, sensitivity and area under curve reached 82.4%, 79.8%, 84.4%, 87.5%, respectively. The segmentation and classification results are both better than other’s methods and mainstream networks. CONCLUSIONS: In order to solve existing problems of liver tumor CT image classification methods, our method realizes the accurate segmentation and classification of liver tumors in CT images and has important clinical application value. Show more
Keywords: Liver CT image, liver tumor segmentation, FLAS-UNet++, liver tumors classification, IDenseNet
DOI: 10.3233/THC-213655
Citation: Technology and Health Care, vol. 30, no. 6, pp. 1475-1487, 2022
Authors: Zhao, Huairui | Hua, Jia | Geng, Xiaochuan | Xu, Jianrong | Guo, Yi | Suo, Shiteng | Zhou, Yan | Wang, Yuanyuan
Article Type: Research Article
Abstract: BACKGROUND: High-precision detection for individual and clustered microcalcifications in mammograms is important for the early diagnosis of breast cancer. Large-scale differences between the two types and low-contrast images are major difficulties faced by radiologists when performing diagnoses. OBJECTIVE: Deep learning-based methods can provide end-to-end solutions for efficient detection. However, multicenter data bias, the low resolution of network inputs, and scale differences between microcalcifications lead to low detection rates. Aiming to overcome the aforementioned limitations, we propose a pyramid feature network for microcalcification detection in mammograms, MicroDMa, with adaptive image adjustment and shortcut connections. METHODS: First, mammograms from multiple centers are …represented as histograms and cropped by adaptive image adjustment, which mitigates the impact of dataset bias. Second, the proposed shortcut connection pyramid network ensures that the feature map contains more information for multiscale objects, while a shortcut path that jumps over layers enhances the efficiency of feature propagation from bottom to top. Third, the weights of each feature map at different scales in the fusion are trainable; thus, the network can automatically learn the contributions of all feature maps in the fusion stage. RESULT: Experiments were conducted on our in-house dataset and the public dataset INbreast. When the average number of positives per image is one on the in-house dataset, the recall rates of MicroDMa are the 96.8% for individual microcalcification and 98.9% for clustered microcalcification, which are higher than 69.1% and 91.2% achieved by recent deep learning model. Free-response receiver operating characteristic curve of MicroDMa is also higher than other methods when models are performed on INbreast. CONCLUSION: MicroDMa network is better than other methods and it can effectively help radiologists detect and identify two types of microcalcifications in clinical applications. Show more
Keywords: Deep learning, detection, pyramid network, mammography, breast microcalcification, convolutional neural network
DOI: 10.3233/THC-220235
Citation: Technology and Health Care, vol. 31, no. 3, pp. 841-853, 2023