Radiology Reports Research Articles

Interactive dual-stream contrastive learning for radiology report generation

Radiology report generation automates diagnostic narrative synthesis from medical imaging data. Current report generation methods primarily employ knowledge graphs for image enhancement, neglecting the interpretability and guiding function of the knowledge graphs themselves. Additionally, few approaches leverage the stable modal alignment information from multimodal pre-trained models to facilitate the generation of radiology reports. We propose the Terms-Guided Radiology Report Generation (TGR), a simple and practical model for generating reports guided primarily by anatomical terms. Specifically, we utilize a dual-stream visual feature extraction module comprised of detail extraction module and a frozen multimodal pre-trained model to separately extract visual detail features and semantic features. Furthermore, a Visual Enhancement Module (VEM) is proposed to further enrich the visual features, thereby facilitating the generation of a list of anatomical terms. We integrate anatomical terms with image features and proceed to engage contrastive learning with frozen text embeddings, utilizing the stable feature space from these embeddings to boost modal alignment capabilities further. Our model incorporates the capability for manual input, enabling it to generate a list of organs for specifically focused abnormal areas or to produce more accurate single-sentence descriptions based on selected anatomical terms. Comprehensive experiments demonstrate the effectiveness of our method in report generation tasks, our TGR-S model reduces training parameters by 38.9% while performing comparably to current state-of-the-art models, and our TGR-B model exceeds the best baseline models across multiple metrics.

Comparative analysis of GPT-4-based ChatGPT's diagnostic performance with radiologists using real-world radiology reports of brain tumors.

Large language models like GPT-4 have demonstrated potential for diagnosis in radiology. Previous studies investigating this potential primarily utilized quizzes from academic journals. This study aimed to assess the diagnostic capabilities of GPT-4-based Chat Generative Pre-trained Transformer (ChatGPT) using actual clinical radiology reports of brain tumors and compare its performance with that of neuroradiologists and general radiologists. We collected brain MRI reports written in Japanese from preoperative brain tumor patients at two institutions from January 2017 to December 2021. The MRI reports were translated into English by radiologists. GPT-4 and five radiologists were presented with the same textual findings from the reports and asked to suggest differential and final diagnoses. The pathological diagnosis of the excised tumor served as the ground truth. McNemar's test and Fisher's exact test were used for statistical analysis. In a study analyzing 150 radiological reports, GPT-4 achieved a final diagnostic accuracy of 73%, while radiologists' accuracy ranged from 65 to 79%. GPT-4's final diagnostic accuracy using reports from neuroradiologists was higher at 80%, compared to 60% using those from general radiologists. In the realm of differential diagnoses, GPT-4's accuracy was 94%, while radiologists' fell between 73 and 89%. Notably, for these differential diagnoses, GPT-4's accuracy remained consistent whether reports were from neuroradiologists or general radiologists. GPT-4 exhibited good diagnostic capability, comparable to neuroradiologists in differentiating brain tumors from MRI reports. GPT-4 can be a second opinion for neuroradiologists on final diagnoses and a guidance tool for general radiologists and residents. This study evaluated GPT-4-based ChatGPT's diagnostic capabilities using real-world clinical MRI reports from brain tumor cases, revealing that its accuracy in interpreting brain tumors from MRI findings is competitive with radiologists. We investigated the diagnostic accuracy of GPT-4 using real-world clinical MRI reports of brain tumors. GPT-4 achieved final and differential diagnostic accuracy that is comparable with neuroradiologists. GPT-4 has the potential to improve the diagnostic process in clinical radiology.

Use of reporting templates for chest radiographs in a coronavirus disease 2019 context: measuring concordance of radiologists with three international templates.

Reporting templates for chest radiographs (CXRs) for patients presenting or being clinically managed for severe acute respiratory syndrome coronavirus 2 [coronavirus disease 2019 (COVID-19)] has attracted advocacy from international radiology societies. We aim to explore the effectiveness and useability of three international templates through the concordance of, and between, radiologists reporting on the presence and severity of COVID-19 on CXRs. Seventy CXRs were obtained from a referral hospital, 50 from patients with COVID-19 (30 rated "classic" COVID-19 appearance and 20 "indeterminate") and 10 "normal" and 10 "alternative pathology" CXRs. The recruited radiologists were assigned to three test sets with the same CXRs but with different template orders. Each radiologist read their test set three times and assigned a classification to the CXR using the Royal Australian New Zealand College of Radiology (RANZCR), British Society of Thoracic Imaging (BSTI), and Modified COVID-19 Reporting and Data System (Dutch; mCO-RADS) templates. Inter-reader variability and intra-reader variability were measured using Fleiss' kappa coefficient. Twelve Australian radiologists participated. The BSTI template had the highest inter-reader agreement (0.46; "moderate" agreement), followed by RANZCR (0.45) and mCO-RADS (0.32). Concordance was driven by strong agreement in "normal" and "alternative" classifications and was lowest for "indeterminate." General consistency was observed across classifications and templates, with intra-reader variability ranging from "good" to "very good" for COVID-19 CXRs (0.61), "normal" CXRs (0.76), and "alternative" (0.68). Reporting templates may be useful in reducing variation among radiology reports, with intra-reader variability showing promise. Feasibility and implementation require a wider approach including referring and treating doctors plus the development of training packages for radiologists specific to the template being used.

Precise Image-level Localization of Intracranial Hemorrhage on Head CT Scans with Deep Learning Models Trained on Study-level Labels.

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To develop a highly generalizable weakly supervised model to automatically detect and localize image- level intracranial hemorrhage (ICH) using study-level labels. Materials and Methods In this retrospective study, the proposed model was pretrained on the image-level RSNA dataset and fine-tuned on a local dataset using attention-based bidirectional long-short-term memory networks. This local training dataset included 10,699 noncontrast head CT scans from 7469 patients with ICH study-level labels extracted from radiology reports. Model performance was compared with that of two senior neuroradiologists on 100 random test scans using the McNemar test, and its generalizability was evaluated on an external independent dataset. Results The model achieved a positive predictive value (PPV) of 85.7% (95% CI: [84.0%, 87.4%]) and an AUC of 0.96 (95% CI: [0.96, 0.97]) on the held-out local test set (n = 7243, 3721 female) and 89.3% (95% CI: [87.8%, 90.7%]) and 0.96 (95% CI: [0.96, 0.97]), respectively, on the external test set (n = 491, 178 female). For 100 randomly selected samples, the model achieved performance on par with two neuroradiologists, but with a significantly faster (P < .05) diagnostic time of 5.04 seconds per scan (versus 86 seconds and 22.2 seconds for the two neuroradiologists, respectively). The model's attention weights and heatmaps visually aligned with neuroradiologists' interpretations. Conclusion The proposed model demonstrated high generalizability and high PPVs, offering a valuable tool for expedited ICH detection and prioritization while reducing false-positive interruptions in radiologists' workflows. ©RSNA, 2024.

Limited Utility of Dynamic Contrast Enhancement Imaging Sequences Within the PI-RADS v2.1 Classification Scheme: A Retrospective Cross-Sectional Study of MRI Reports.

Background/Objective: We sought to characterize the proportion of peripheral zone lesions "upgraded" within the PI-RADS v2.1 protocol using DCE imaging sequences in a large patient population undergoing multiparametric prostate MRI. Methods: A retrospective review of radiologist reports for 2742 prostate MRI exams at 2 large Alberta teaching hospitals between January 2017 and January 2022 was conducted. Prostate specific antigen (PSA), prostate volume, sequence specific and overall PI-RADS scores, and lesion positivity for DCE were collected if present in the accompanying radiology report. Further, pathology reports of biopsies of the upgraded lesions within upgraded patients were reviewed to see if upgraded lesions were deemed clinically significant by gleason score/grade group. Results: The median age was 63 years, with a median PSA and PSA density of 7.5 ng/mL and 0.13 ng/mL2 respectively. A total of 1809 lesions were reported, with 69.4% of all lesions being DCE positive. Of the lesions within the peripheral zone, 548 were overall PI-RADS 4. A total of 87/2742 (3.2%) of patients were upgraded to a PI-RADS 4 by DCE imaging. Within these patients, 65 had pathology reports available, of which 18 had a clinically significant lesion at the upgrade site. Conclusion: Contrast enhancement is only beneficial for a very small portion of patients undergoing prostate MRI. Given the invasive nature of contrast enhanced studies, potential contrast induced side effects, added imaging time, and the cost of contrast agent, routine use of contrast for prostate MRI is questioned. Further studies are necessary to determine if it should be part of routine prostate MRI imaging protocols.

Spleen anomalies and lesions in CT and MRI: essentials for radiologists and clinicians-a pictorial review.

The precise evaluation of splenic lesions using cross-sectional imaging presents a complex challenge due to the overlapping imaging characteristics of most splenic diseases and the absence of specific symptoms, complicating the diagnosis process. This pictorial review highlights the essential elements that should be included in a radiology report to interpret cross-sectional imaging studies of splenic lesions. To augment the diagnostic accuracy, we propose the implementation of a stepwise algorithmic approach. This approach offers a comprehensive analysis of splenic anomalies and lesions, providing an invaluable tool for radiologists and clinicians.

The Fine-Tuned Large Language Model for Extracting the Progressive Bone Metastasis from Unstructured Radiology Reports.

Early detection of patients with impending bone metastasis is crucial for prognosis improvement. This study aimed to investigate the feasibility of a fine-tuned, locally run large language model (LLM) in extracting patients with bone metastasis in unstructured Japanese radiology report and to compare its performance with manual annotation. This retrospective study included patients with "metastasis" in radiological reports (April 2018-January 2019, August-May 2022, and April-December 2023 for training, validation, and test datasets of 9559, 1498, and 7399 patients, respectively). Radiologists reviewed the clinical indication and diagnosis sections of the radiological report (used as input data) and classified them into groups 0 (no bone metastasis), 1 (progressive bone metastasis), and 2 (stable or decreased bone metastasis). The data for group 0 was under-sampled in training and test datasets due to group imbalance. The best-performing model from the validation set was subsequently tested using the testing dataset. Two additional radiologists (readers 1 and 2) were involved in classifying radiological reports within the test dataset for testing purposes. The fine-tuned LLM, reader 1, and reader 2 demonstrated an accuracy of 0.979, 0.996, and 0.993, sensitivity for groups 0/1/2 of 0.988/0.947/0.943, 1.000/1.000/0.966, and 1.000/0.982/0.954, and time required for classification (s) of 105, 2312, and 3094 in under-sampled test dataset (n = 711), respectively. Fine-tuned LLM extracted patients with bone metastasis, demonstrating satisfactory performance that was comparable to or slightly lower than manual annotation by radiologists in a noticeably shorter time.

Preliminary image evaluation performance of radiographers in one New Zealand District: a 6-month prospective study.

Preliminary image evaluation (PIE) is a system where radiographers alert emergency department referrers to the presence or absence of abnormalities on acute extremity X-ray examinations. PIE and similar systems have been utilised in the United Kingdom (UK) and Australia due to a shortage of radiologists to provide a timely report. As New Zealand (NZ) faces a similar shortage, PIE should be considered to address the negative impact this has on patients. The aim of this study was to determine the effect of regular feedback and education on radiographers' performance when detecting and describing acute abnormalities on extremity X-ray examinations in ED. A prospective longitudinal study design was utilised for this study. The sensitivity, specificity, accuracy, and accuracy of PIEs performed by seven radiographers at a public provincial district in NZ were assessed over a 6-month period, with the participants provided monthly results along with regular e-mailed feedback on common errors. The mean for sensitivity, specificity, and accuracy calculated with a 95% confidence interval over the 6-month period were 92.8% (89.9, 95.8), 94.9 (93.1, 96.8), and 94.2 (91.9, 96.5), respectively. When the month-to-month results were analysed, the results demonstrated an improvement in participants' sensitivity, specificity, and accuracy over the 6-month period. The results of this study demonstrated that radiographers who participated in the study can perform PIE to a high standard that is comparable with the findings from international studies and demonstrated an improvement over 6 months. Therefore, PIE may be useful in NZ to aid ED clinicians in their clinical decisions when a radiology report is unavailable.

Establishment of a Best Practices in Reporting (BPR) Recommendation for Abdominal Aortic Aneurysms in a Large Multi-State Radiology Practice: Adoption and Impact

Purpose of the study: To evaluate the performance of Best Practice Recommendation (BPR) compliance in reporting abdominal aortic aneurysm findings on imaging, comparing the results before and after its deployment. Methods: Best Practice Recommendations for AAA were deployed in 2020 at a large radiology practice site. Reports between January 2018 through October 2022 were reviewed, representing studies read prior to and subsequent to the implementation of the reporting standards. Cases of abdominal aortic aneurysms ≥ 2.6 cm were counted by year. Adherence to the BPR for each year was calculated as [total number of confirmed cases of ≥ 2.6 cm AAAs with compliant reports] * 100 / [the total number of confirmed ≥ 2.6 cm AAAs]. A secondary analysis was performed to determine whether there was a statistically significant difference in the proportion of BPR-compliant reports for AAA cases before (from 2018 to 2019) and after (from 2020 to 2022) BPR deployment using a chi-square test. Results: From January 2018 to December 2022, there were 8,693 reports referencing AAA. After excluding cases of suspected AAA (N = 2,131), confirmed AAAs with indeterminate sizes (N = 103), and confirmed AAAs with sizes &lt; 2.6 cm (N = 85), the number of AAA cases ≥ 2.6 cm in size was 6,374. Concordance with the BPR standards for the remaining cases with sizes ≥ 2.6 cm were 1.6% and 4.1% in 2018 and 2019, respectively. Post-implementation of BPRs, there was a substantial improvement in guideline adherence to 32.1%, 84.3%, and 83.6% in 2020, 2021, and 2022, respectively. In general, the proportion of BPR-compliant reports of AAA cases in the pre-deployment (3.6%) period statistically differs (p - value &lt; 0.0001) from those in the post-deployment period (73.9%) Conclusion: Adherence to reporting standards increased after the BPR deployment in 2020. The inclusion of management recommendations in the radiology report when AAA is identified is a simple and cost-effective way of improving outcomes for patients with AAAs through appropriate follow-up treatment.

A vision-language foundation model for the generation of realistic chest X-ray images.

The paucity of high-quality medical imaging datasets could be mitigated by machine learning models that generate compositionally diverse images that faithfully represent medical concepts and pathologies. However, large vision-language models are trained on natural images, and the diversity distribution of the generated images substantially differs from that of medical images. Moreover, medical language involves specific and semantically rich vocabulary. Here we describe a domain-adaptation strategy for large vision-language models that overcomes distributional shifts. Specifically, by leveraging publicly available datasets of chest X-ray images and the corresponding radiology reports, we adapted a latent diffusion model pre-trained on pairs of natural images and text descriptors to generate diverse and visually plausible synthetic chest X-ray images (as confirmed by board-certified radiologists) whose appearance can be controlled with free-form medical text prompts. The domain-adaptation strategy for the text-conditioned synthesis of medical images can be used to augment training datasets and is a viable alternative to the sharing of real medical images for model training and fine-tuning.

A scoping review of large language model based approaches for information extraction from radiology reports

Radiological imaging is a globally prevalent diagnostic method, yet the free text contained in radiology reports is not frequently used for secondary purposes. Natural Language Processing can provide structured data retrieved from these reports. This paper provides a summary of the current state of research on Large Language Model (LLM) based approaches for information extraction (IE) from radiology reports. We conduct a scoping review that follows the PRISMA-ScR guideline. Queries of five databases were conducted on August 1st 2023. Among the 34 studies that met inclusion criteria, only pre-transformer and encoder-based models are described. External validation shows a general performance decrease, although LLMs might improve generalizability of IE approaches. Reports related to CT and MRI examinations, as well as thoracic reports, prevail. Most common challenges reported are missing validation on external data and augmentation of the described methods. Different reporting granularities affect the comparability and transparency of approaches.

Generating colloquial radiology reports with large language models.

Patients are increasingly being given direct access to their medical records. However, radiology reports are written for clinicians and typically contain medical jargon, which can be confusing. One solution is for radiologists to provide a "colloquial" version that is accessible to the layperson. Because manually generating these colloquial translations would represent a significant burden for radiologists, a way to automatically produce accurate, accessible patient-facing reports is desired. We propose a novel method to produce colloquial translations of radiology reports by providing specialized prompts to a large language model (LLM). Our method automatically extracts and defines medical terms and includes their definitions in the LLM prompt. Using our method and a naive strategy, translations were generated at 4 different reading levels for 100 de-identified neuroradiology reports from an academic medical center. Translations were evaluated by a panel of radiologists for accuracy, likability, harm potential, and readability. Our approach translated the Findings and Impression sections at the 8th-grade level with accuracies of 88% and 93%, respectively. Across all grade levels, our approach was 20% more accurate than the baseline method. Overall, translations were more readable than the original reports, as evaluated using standard readability indices. We find that our translations at the eighth-grade level strike an optimal balance between accuracy and readability. Notably, this corresponds to nationally recognized recommendations for patient-facing health communication. We believe that using this approach to draft patient-accessible reports will benefit patients without significantly increasing the burden on radiologists.

Assessment of Follow-Up for Pulmonary Nodules from Radiology Reports with Natural Language Processing.

Radiology reports are an essential communication method for ensuring smooth workflow in healthcare. However, many of these reports are described in free text, and findings documented by radiologists may not be adequately addressed. In this study, focusing on pulmonary nodules, we evaluated whether cases in which radiologists described follow-up as recommended were receiving appropriate treatment. Reports recommending follow-up for pulmonary nodules were automatically extracted using natural language processing. In our evaluation, out of 10,507 reports, 1,501 cases (14.3%) were classified as "reports recommending follow-up for pulmonary nodules." Among these, 958 cases underwent additional imaging tests within 400 days. From the remaining 543 cases, we randomly sampled 42 cases and conducted chart reviews by clinicians to confirm patient care status. Our assessment found that follow-up was not documented in 17 of the 42 cases (40.5%), indicating a high likelihood that appropriate care was not provided.

ChatGPT in Radiology: Insights into Current Advantages and Limitations of Artificial Intelligence in Radiology Reporting

ChatGPT in Radiology: Insights into Current Advantages and Limitations of Artificial Intelligence in Radiology Reporting

A Novel Deep Learning Approach for the Automatic Diagnosis of Acute Appendicitis.

Background: Acute appendicitis (AA) is a major cause of acute abdominal pain requiring surgical intervention. Approximately 20% of AA cases are diagnosed neither early nor accurately, leading to an increased risk of appendiceal perforation and postoperative sequelae. AA can be identified with good accuracy using computed tomography (CT). However, some studies have found that a false-negative AA diagnosis made using CT can cause surgical therapy to be delayed. Deep learning experiments are aimed at minimizing false-negative diagnoses. However, the success rates reported in these studies are far from 100%. In addition, the methods used to divide patients into groups do not adequately reflect situations in which accurate radiological diagnosis is difficult. Therefore, in this study, we propose a novel deep-learning approach for the automatic diagnosis of AA using CT based on establishing a new strategy for classification according to the difficulties encountered in radiological diagnosis. Methods: A total of 266 patients with a pathological diagnosis of AA who underwent appendectomy were divided into two groups based on CT images and radiology reports. A deep learning analysis was performed on the CT images and clinical and laboratory parameters that contributed to the diagnosis of both the patient and age- and sex-adjusted control groups. Results: The deep learning diagnosis success rate was 96% for the group with advanced radiological findings and 83.3% for the group with radiologically suspicious findings that could be considered normal. Conclusions: Using deep learning, successful results can be achieved in cases in which the appendix diameter has not increased significantly and there is no significant edema effect.

Staging peritoneal metastases in colorectal cancer: The correlation between MRI, surgical and histopathological peritoneal cancer index

IntroductionDW-MRI is a non-invasive way to determine the peritoneal cancer index (PCI) in colorectal cancer (CRC) patients with peritoneal metastases (PM). However, like surgeons during surgery, radiologists struggle to differentiate between PM and fibrosis. This study aimed to investigate the agreement between the PCI as determined by MRI (mriPCI), during surgery (sPCI) and histopathology examination (pPCI) in CRC patients undergoing cytoreductive surgery and hyperthermic intraperitoneal chemotherapy (CRS-HIPEC). Materials and methodsThis was a single-centre, retrospective study of CRC patients with PM who were staged with DW-MRI and underwent subsequent CRS-HIPEC. All initial patients' radiological, surgical and histopathology reports were reviewed for the PCI. Histopathology was the reference standard. Primary outcome was the correlation and agreement between mriPCI and pPCI. ResultsEighty-seven patients were included. All patients had a complete macroscopic resection. Median (interquartile range) PCI for MRI, surgery, and histopathology were respectively 6.0 (2.5-9.0), 6.0 (4.0-11.0) and 6.0 (2.5-9.5). The intraclass correlation coefficient between the sPCI and pPCI was excellent 0.87 (p <0.001), and good between mriPCI and pPCI 0.77 (p <0.001) and between sPCI and mriPCI 0.70 (p <0.001). ConclusionMRI is a promising non-invasive tool to assess the PCI rather accurately.

Towards a Reporting Guideline for Studies on Information Extraction from Clinical Texts.

The rapid technical progress in the domain of clinical Natural Language Processing and information extraction (IE) has resulted in challenges concerning the comparability and replicability of studies. This paper proposes a reporting guideline to standardize the description of methodologies and outcomes for studies involving IE from clinical texts. The guideline is developed based on the experiences gained from data extraction for a previously conducted scoping review on IE from free-text radiology reports including 34 studies. The guideline comprises the five top-level categories information model, architecture, data, annotation, and outcomes. In total, we define 28 aspects to be reported on in IE studies related to these categories. The proposed guideline is expected to set a standard for reporting in studies describing IE from clinical text and promote uniformity across the research field. Expected future technological advancements may make regular updates of the guideline necessary. In future research, we plan to develop a taxonomy that clearly defines corresponding value sets as well as integrating both this guideline and the taxonomy by following a consensus-based methodology.

Collaborating with non-radiological clinical colleagues.

Paediatric radiology is a challenging and intriguing subspecialty, dealing with children, guardians and non-radiological clinical colleagues. Paediatric radiologists are routinely in contact with numerous paediatric and surgical subspecialties, all having different needs, perceptions, prioritisations and expectations. Moreover, the radiologist is part of the team of radiographers, sonographers, nurses and secretaries, assisted by appropriate equipment and electronic tools. The framework of good collaboration to ensure safety and effectiveness for the imaged child is a shared responsibility among all medical practitioners involved. Communication in routine practice has many forms and includes appropriately filled radiology requests in accordance to the patient's medical records, routine and timely production of structured, problem-solving radiology reports, face-to-face or electronic-assisted communications and discussions on a pre-defined framework, mutually-agreed and evidence-based protocols adjusted to local availability, skills and national and international guidelines. Mutual understanding of advantages and limitations of imaging is paramount. Well-meant discussions, professionalism and empathy should promote soft skills, bidirectional communication and good collaboration for the benefit of added-value paediatric radiology. International societies, health authorities, medical directors and senior consultants have the responsibility to suggest and safeguard frameworks and recommendations. Regular multidisciplinary meetings and multidisciplinary research projects under openness, honesty and transparency are pathways favouring good collaboration.

Preliminary assessment of TNM classification performance for pancreatic cancer in Japanese radiology reports using GPT-4.

A large-scale language model is expected to have been trained with a large volume of data including cancer treatment protocols. The current study aimed to investigate the use of generative pretrained transformer 4 (GPT-4) for identifying the TNM classification of pancreatic cancers from existing radiology reports written in Japanese. We screened 100 consecutive radiology reports on computed tomography scan for pancreatic cancer from April 2020 to June 2022. GPT-4 was requested to classify the TNM from the radiology reports based on the General Rules for the Study of Pancreatic Cancer 7th Edition. The accuracy and kappa coefficient of the TNM classifications by GPT-4 was evaluated with the classifications by two experienced abdominal radiologists as gold standard. The accuracy values of the T, N, and M factors were 0.73, 0.91, and 0.93, respectively. The kappa coefficients were 0.45 for T, 0.79 for N, and 0.83 for M. Although GPT is familiar with the TNM classification for pancreatic cancer, its performance in classifying actual cases in this experiment may not be adequate.

DKA-RG: Disease-Knowledge-Enhanced Fine-Grained Image–Text Alignment for Automatic Radiology Report Generation

Automatic radiology report generation is a task that combines artificial intelligence and medical information processing, and it fully relies on computer vision and natural language processing techniques. Nowadays, automatic radiology report generation is still a very challenging task because it requires semantically adequate alignment of data from two modalities: radiology images and text. Existing approaches tend to focus on coarse-grained alignment at the global level and do not take into account the disease characteristics of radiology images at fine-grained semantics, which results in the generated reports potentially omitting key disease diagnostic descriptions. In this work, we propose a new approach, disease-knowledge-enhanced fine-grained image–text alignment for automatic radiology report generation (DKA-RG). The method combines global and disease-level alignment, thus facilitating the extraction of fine-grained disease features by the model. Our approach also introduces a knowledge graph to inject medical domain expertise into the model. Our proposed DKA-RG consists of two training steps: the image–report alignment stage and the image-to-report generation stage. In the alignment stage, we use global contrastive learning to align images and texts from a high level and also augment disease contrastive learning with medical knowledge to enhance the disease detection capability. In the report generation stage, the report text generated from the images is more accurate in describing the disease information thanks to sufficient alignment. Through extensive quantitative and qualitative experiments on two widely used datasets, we validate the effectiveness of our DKA-RG on the task of radiology report generation. Our DKA-RG achieves superior performance on multiple types of metrics (natural language generation and clinical efficacy metrics) compared to existing methods, demonstrating that the method can improve the reliability and accuracy of automatic radiology report generation systems.