Search

Purpose: While dual-energy computed tomography (DECT) images provide clinically useful information than single-energy CT (SECT), SECT remains the most widely used CT system globally, and only a few institutions can use DECT. This study aimed to establish an artificial intelligence (AI)-based image-domain material decomposition technique using multiple keV-output learning of virtual monochromatic images (VMIs) to create DECT-equivalent images from SECT images., Methods: This study involved 82 patients with head and neck cancer. Of these, the AI model was built with data from the 67 patients with only DECT scans, while 15 patients with both SECT and DECT scans were used for SECT testing. Our AI model generated VMI 50keV and VMI 100keV from VMI 70keV equivalent to 120-kVp SECT images. We introduced a loss function for material density images (MDIs) in addition to the loss for VMIs. For comparison, we trained the same model with the loss for VMIs only. DECT-equivalent images were generated from SECT images and compared with the true DECT images., Results: The prediction time was 5.4 s per patient. The proposed method with the MDI loss function quantitatively provided more accurate DECT-equivalent images than the model trained with the loss for VMIs only. Using real 120-kVp SECT images, the trained model produced precise DECT images of excellent quality., Conclusion: In this study, we developed an AI-based material decomposition approach for head and neck cancer patients by introducing the loss function for MDIs via multiple keV-output learning. Our results suggest the feasibility of AI-based image-domain material decomposition in a conventional SECT system without a DECT scanner., (© 2024. CARS.)

Purpose: Spinal bone metastases directly affect quality of life, and patients with lytic-dominant lesions are at high risk for neurological symptoms and fractures. To detect and classify lytic spinal bone metastasis using routine computed tomography (CT) scans, we developed a deep learning (DL)-based computer-aided detection (CAD) system., Methods: We retrospectively analyzed 2125 diagnostic and radiotherapeutic CT images of 79 patients. Images annotated as tumor (positive) or not (negative) were randomized into training (1782 images) and test (343 images) datasets. YOLOv5m architecture was used to detect vertebra on whole CT scans. InceptionV3 architecture with the transfer-learning technique was used to classify the presence/absence of lytic lesions on CT images showing the presence of vertebra. The DL models were evaluated via fivefold cross-validation. For vertebra detection, bounding box accuracy was estimated using intersection over union (IoU). We evaluated the area under the curve (AUC) of a receiver operating characteristic curve to classify lesions. Moreover, we determined the accuracy, precision, recall, and F1 score. We used the gradient-weighted class activation mapping (Grad-CAM) technique for visual interpretation., Results: The computation time was 0.44 s per image. The average IoU value of the predicted vertebra was 0.923 ± 0.052 (0.684-1.000) for test datasets. In the binary classification task, the accuracy, precision, recall, F1-score, and AUC value for test datasets were 0.872, 0.948, 0.741, 0.832, and 0.941, respectively. Heat maps constructed using the Grad-CAM technique were consistent with the location of lytic lesions., Conclusion: Our artificial intelligence-aided CAD system using two DL models could rapidly identify vertebra bone from whole CT images and detect lytic spinal bone metastasis, although further evaluation of diagnostic accuracy is required with a larger sample size., (© 2023. CARS.)

Purpose: Deep learning (DL) is a state-of-the-art technique for developing artificial intelligence in various domains and it improves the performance of natural language processing (NLP). Therefore, we aimed to develop a DL-based NLP model that classifies the status of bone metastasis (BM) in radiology reports to detect patients with BM., Materials and Methods: The DL-based NLP model was developed by training long short-term memory using 1,749 free-text radiology reports written in Japanese. We adopted five-fold cross-validation and used 200 reports for testing the five models. The accuracy, sensitivity, specificity, precision, and area under the receiver operating characteristics curve (AUROC) were used for the model evaluation., Results: The developed model demonstrated classification performance with mean ± standard deviation of 0.912 ± 0.012, 0.924 ± 0.029, 0.901 ± 0.014, 0.898 ± 0.012, and 0.968 ± 0.004 for accuracy, sensitivity, specificity, precision, and AUROC, respectively., Conclusion: The proposed DL-based NLP model may help in the early and efficient detection of patients with BM., (© 2023. The Author(s) under exclusive licence to Japan Radiological Society.)

Purpose: Low-energy virtual monochromatic images (VMIs) derived from dual-energy computed tomography (DECT) systems improve lesion conspicuity of head and neck cancer over single-energy CT (SECT). However, DECT systems are installed in a limited number of facilities; thus, only a few facilities benefit from VMIs. In this work, we present a deep learning (DL) architecture suitable for generating pseudo low-energy VMIs of head and neck cancers for facilities that employ SECT imaging., Methods: We retrospectively analyzed 115 patients with head and neck cancers who underwent contrast enhanced DECT. VMIs at 70 and 50 keV were used as the input and ground truth (GT), respectively. We divided them into two datasets: for DL (104 patients) and for inference with SECT (11 patients). We compared four DL architectures: U-Net, DenseNet-based, and two ResNet-based models. Pseudo VMIs at 50 keV (pVMI 50keV ) were compared with the GT in terms of the mean absolute error (MAE) of Hounsfield unit (HU) values, peak signal-to-noise ratio (PSNR), and structural similarity (SSIM). The HU values for tumors, vessels, parotid glands, muscle, fat, and bone were evaluated. pVMI 50keV were generated from actual SECT images and the HU values were evaluated., Results: U-Net produced the lowest MAE (13.32 ± 2.20 HU) and highest PSNR (47.03 ± 2.33 dB) and SSIM (0.9965 ± 0.0009), with statistically significant differences (P < 0.001). The HU evaluation showed good agreement between the GT and U-Net. U-Net produced the smallest absolute HU difference for the tumor, at < 5.0 HU., Conclusion: Quantitative comparisons of physical parameters demonstrated that the proposed U-Net could generate high accuracy pVMI 50keV in a shorter time compared with the established DL architectures. Although further evaluation on diagnostic accuracy is required, our method can help obtain low-energy VMI from SECT images without DECT systems., (© 2022. CARS.)