To assess whether transfer learning with a bidirectional encoder representations from transformers (BERT) model, pretrained on a clinical corpus, can perform sentence-level anatomic classification of free-text radiology reports, even for anatomic classes with few positive examples. This retrospective study included radiology reports of patients who underwent whole-body PET/CT imaging from December 2005 to December 2020. Each sentence in these reports (6272 sentences) was labeled by two annotators according to body part (“brain,” “head & neck,” “chest,” “abdomen,” “limbs,” “spine,” or “others”). The BERT-based transfer learning approach was compared with two baseline machine learning approaches: bidirectional long short-term memory (BiLSTM) and the count-based method. Area under the precision-recall curve (AUPRC) and area under the receiver operating characteristic curve (AUC) were computed for each approach, and AUCs were compared using the DeLong test. The BERT-based approach achieved a macro-averaged AUPRC of 0.88 for classification, outperforming the baselines. AUC results for BERT were significantly higher than those of BiLSTM for all classes and those of the count-based method for the “brain,” “chest,” “abdomen,” and “others” classes (P values < .025). AUPRC results for BERT were superior to those of baselines even for classes with few labeled training data (brain: BERT, 0.95, BiLSTM, 0.11, count based, 0.41; limbs: BERT, 0.74, BiLSTM, 0.28, count based, 0.46; spine: BERT, 0.82, BiLSTM, 0.53, count based, 0.69). The BERT-based transfer learning approach outperformed the BiLSTM and count-based approaches in sentence-level anatomic classification of free-text radiology reports, even for anatomic classes with few labeled training data. Keywords: Anatomy, Comparative Studies, Technology Assessment, Transfer Learning Supplemental material is available for this article. © RSNA, 2023

Liver cancer remains a significant global health concern, ranking as the sixth most common malignancy and the third leading cause of cancer-related deaths worldwide. Medical imaging plays a vital role in managing liver tumors, particularly hepatocellular carcinoma (HCC) and metastatic lesions. However, the large volume and complexity of imaging data can make accurate and efficient interpretation challenging. Artificial intelligence (AI) is recognized as a promising tool to address these challenges. Therefore, this review aims to explore the recent advances in AI applications in liver tumor imaging, focusing on key areas such as image reconstruction, image quality enhancement, lesion detection, tumor characterization, segmentation, and radiomics. Among these, AI-based image reconstruction has already been widely integrated into clinical workflows, helping to enhance image quality while reducing radiation exposure. While the adoption of AI-assisted diagnostic tools in liver imaging has lagged behind other fields, such as chest imaging, recent developments are driving their increasing integration into clinical practice. In the future, AI is expected to play a central role in various aspects of liver cancer care, including comprehensive image analysis, treatment planning, response evaluation, and prognosis prediction. This review offers a comprehensive overview of the status and prospects of AI applications in liver tumor imaging.

Artificial intelligence (AI) systems designed to detect abnormalities in abdominal computed tomography (CT) could reduce radiologists' workload and improve diagnostic processes. However, development of such models has been hampered by the shortage of large expert-annotated datasets. Here, we used information from free-text radiology reports, rather than manual annotations, to develop a deep-learning-based pipeline for comprehensive detection of abdominal CT abnormalities. In this multicentre retrospective study, we developed a deep-learning-based pipeline to detect abnormalities in the liver, gallbladder, pancreas, spleen, and kidneys. Abdominal CT exams and related free-text reports obtained during routine clinical practice collected from three institutions were used for training and internal testing, while data collected from six institutions were used for external testing. A multi-organ segmentation model and an information extraction schema were used to extract specific organ images and disease information, CT images and radiology reports, respectively, which were used to train a multiple-instance learning model for anomaly detection. Its performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, and F1 score against radiologists' ground-truth labels. We trained the model for each organ on images selected from 66,684 exams (39,255 patients) and tested it on 300 (295 patients) and 600 (596 patients) exams for internal and external validation, respectively. In the external test cohort, the overall AUC for detecting organ abnormalities was 0.886. Whereas models trained on human-annotated labels performed better with the same number of exams, those trained on larger datasets with labels auto-extracted via the information extraction schema significantly outperformed human-annotated label-derived models. Using disease information from routine clinical free-text radiology reports allows development of accurate anomaly detection models without requiring manual annotations. This approach is applicable to various anatomical sites and could streamline diagnostic processes. Japan Science and Technology Agency.

Purpose: To propose a five-point scale for radiology report importance called Report Importance Category (RIC) and to compare the performance of natural language processing (NLP) algorithms in assessing RIC using head computed tomography (CT) reports written in Japanese. Materials and methods: 3728 Japanese head CT reports performed at Osaka University Hospital in 2020 were included. RIC (category 0: no findings, category 1: minor findings, category 2: routine follow-up, category 3: careful follow-up, and category 4: examination or therapy) was established based not only on patient severity but also on the novelty of the information. The manual assessment of RIC for the reports was performed under the consensus of two out of four neuroradiologists. The performance of four NLP models for classifying RIC was compared using fivefold cross-validation: logistic regression, bidirectional long–short-term memory (BiLSTM), general bidirectional encoder representations of transformers (general BERT), and domain-specific BERT (BERT for medical domain). Results: The proportion of each RIC in the whole data set was 15.0%, 26.7%, 44.2%, 7.7%, and 6.4%, respectively. Domain-specific BERT showed the highest accuracy (0.8434 ± 0.0063) in assessing RIC and significantly higher AUC in categories 1 (0.9813 ± 0.0011), 2 (0.9492 ± 0.0045), 3 (0.9637 ± 0.0050), and 4 (0.9548 ± 0.0074) than the other models (p < .05). Analysis using layer-integrated gradients showed that the domain-specific BERT model could detect important words, such as disease names in reports. Conclusions: Domain-specific BERT has superiority over the other models in assessing our newly proposed criteria called RIC of head CT radiology reports. The accumulation of similar and further studies of has a potential to contribute to medical safety by preventing missed important findings by clinicians.