Chest Computed Tomography Reports Research Articles

Exploring Multilingual Large Language Models for Enhanced TNM Classification of Radiology Report in Lung Cancer Staging.

This study aimed to investigate the accuracy of Tumor, Node, Metastasis (TNM) classification based on radiology reports using GPT3.5-turbo (GPT3.5) and the utility of multilingual large language models (LLMs) in both Japanese and English. Utilizing GPT3.5, we developed a system to automatically generate TNM classifications from chest computed tomography reports for lung cancer and evaluate its performance. We statistically analyzed the impact of providing full or partial TNM definitions in both languages using a generalized linear mixed model. The highest accuracy was attained with full TNM definitions and radiology reports in English (M = 94%, N = 80%, T = 47%, and TNM combined = 36%). Providing definitions for each of the T, N, and M factors statistically improved their respective accuracies (T: odds ratio [OR] = 2.35, p < 0.001; N: OR = 1.94, p < 0.01; M: OR = 2.50, p < 0.001). Japanese reports exhibited decreased N and M accuracies (N accuracy: OR = 0.74 and M accuracy: OR = 0.21). This study underscores the potential of multilingual LLMs for automatic TNM classification in radiology reports. Even without additional model training, performance improvements were evident with the provided TNM definitions, indicating LLMs' relevance in radiology contexts.

Chest computed tomography findings among adult Aboriginal Australians with bronchiectasis in the Top End Northern Territory of Australia.

There is limited evidence in the literature illustrating chest computed tomography (CT) characteristics among adult Aboriginal Australians with bronchiectasis. This retrospective study evaluates the radiological characteristics of bronchiectasis in Aboriginal Australians residing in the Top End, Northern Territory of Australia. Patients aged >18 years with chest CT-confirmed bronchiectasis between 2011 and 2020 were included. Demographics and relevant clinical parameters were collected. Alongside confirming bronchiectasis, chest CT reports were assessed for (i) lobar location (ii) unilateral or bilateral involvement and (iii) bronchiectasis type when available. A total of 459 patients were identified with chest CT-confirmed bronchiectasis, with a median age of 47 years, and 55% were females. Bronchiectasis was predominantly recorded in the left lower lobe (LLL) (73%), followed by the right lower lobe (RLL) (62%) and the left upper lobe (LUL) was least common (22%). Females recorded the right middle lobe (RML) affected significantly more often than males (50 vs. 34%, P = 0.012). Bilateral involvement was common (74%), with the strongest pairwise correlation associated between the right upper lobe (RUL) and LUL (P < 0.001). Cylindrical (50%) and cystic (28%) types were most common. The RML and LLL showed positive correlation with cylindrical and LUL with cystic bronchiectasis. Neither lobar location nor bronchiectasis type showed any significant association with lung function parameters other than RML, Lingula and LUL involvement being associated with better percent predicted values of diffusing capacity for carbon monoxide. There were no significant associations between sputum culture and type or lobar locations of bronchiectasis except for non-Aspergillus fungus culture prevalence was higher with cystic or cylindrical types. The results of this study may be an avenue to develop CT bronchiectasis severity scale in the future specific for Aboriginal Australians.

A validation of register-derived diagnoses of interstitial lung disease in patients with inflammatory arthritis: data from the NOR-DMARD study

Objective There is a lack of knowledge concerning the validity of the interstitial lung disease (ILD) diagnoses used in epidemiological studies on rheumatic diseases. This paper seeks to verify register-derived ILD diagnoses using chest computed tomography (CT) and medical records as a gold standard. Method The Norwegian Anti-Rheumatic Drug Register (NOR-DMARD) is a multicentre prospective observational study of patients with inflammatory arthritis who start treatment with disease-modifying anti-rheumatic drugs. NOR-DMARD is linked to the Norwegian Patient Registry (NPR) and Cause of Death Registry. We searched registers for ILD coded by ICD-10 J84 or J99 among patients with rheumatoid arthritis, psoriatic arthritis, or spondyloarthritis. We extracted chest CT reports and medical records from participating hospitals. Two expert thoracic radiologists scored examinations to confirm the ILD diagnosis. We also searched medical records to find justifications for the diagnosis following multidisciplinary evaluations. We calculated the positive predictive values (PPVs) for ILD across subsets. Results We identified 71 cases with an ILD diagnosis. CT examinations were available in 65/71 patients (91.5%), of whom ILD was confirmed on CT in 29/65 (44.6%). In a further 10 patients, medical records confirmed the diagnosis, giving a total of 39/71 verified cases. The PPV of a register-derived ILD diagnosis was thus 54.9%. In a subset of patients who had received an ILD code at two or more time-points and had a CT scan taken within a relevant period, the PPV was 72.2%. Conclusion The validity of register-based diagnoses of ILD must be carefully considered in epidemiological studies.

Classification of Diagnostic Certainty in Radiology Reports with Deep Learning.

A radiology report is prepared for communicating clinical information about observed abnormal structures and clinically important findings with referring clinicians. However, such observations and findings are often accompanied by ambiguous expressions, which can prevent clinicians from accurately interpreting the content of reports. To systematically assess the degree of diagnostic certainty for each observation and finding in a report, we defined an ordinal scale comprising five classes: definite, likely, may represent, unlikely, and denial. Furthermore, we applied a deep learning classification model to determine its applicability to in-house radiology reports. We trained and evaluated the model using 540 in-house chest computed tomography reports. The deep learning model achieved a micro F1-score of 97.61%, which indicated that our ordinal scale was suitable for measuring the diagnostic certainty of observations and findings in a report.

Risk and Time to Diagnosis of Lung Cancer in Incidental Pulmonary Nodules.

To determine the risk of lung cancer in incidental pulmonary nodules, as well as the time until cancer growth is detected. This retrospective study examined patients with incidental nodules detected on chest computed tomography (CT) in 2017. Characteristics of the dominant nodule were automatically extracted from CT reports, and cancer diagnoses were manually verified by a thoracic radiologist. Nodules were categorized per Fleischner Society guideline categories: solid <6mm, solid 6 to 8mm, solid >8mm, subsolid <6mm, ground glass nodules ≥6mm, and part-solid nodules ≥6mm. The time to nodule growth was determined by CT reports. A total of 3180 patients (nodules) were included, of which 155 (5%) were diagnosed with lung cancer. By category, 7/1601 (0.4%) solid nodules <6mm, 11/713 (1.5%) solid nodules 6 to 8mm, 71/446 (15.9%) solid nodules >8mm, 1/124 (0.8%) subsolid nodules <6mm, 29/202 (14.4%) ground glass nodules ≥6mm, and 36/94 (37.9%) part-solid nodules ≥6mm were malignant. Of solid lung cancers <6mm, growth was observed in 1/4 imaged by 1 year and 2/5 by 2 years; of solid lung cancers 6 to 8mm, growth was observed in 3/10 imaged by 1 year and 6/10 by 2 years. Solid nodules <6mm have a very low risk of malignancy and may not require routine follow-up. However, when malignant, growth is often not observed until 2 or more years later; therefore, stability at 1 to 2 years does not imply benignity.

Extracting Clinical Information From Japanese Radiology Reports Using a 2-Stage Deep Learning Approach: Algorithm Development and Validation.

Radiology reports are usually written in a free-text format, which makes it challenging to reuse the reports. For secondary use, we developed a 2-stage deep learning system for extracting clinical information and converting it into a structured format. Our system mainly consists of 2 deep learning modules: entity extraction and relation extraction. For each module, state-of-the-art deep learning models were applied. We trained and evaluated the models using 1040 in-house Japanese computed tomography (CT) reports annotated by medical experts. We also evaluated the performance of the entire pipeline of our system. In addition, the ratio of annotated entities in the reports was measured to validate the coverage of the clinical information with our information model. The microaveraged F1-scores of our best-performing model for entity extraction and relation extraction were 96.1% and 97.4%, respectively. The microaveraged F1-score of the 2-stage system, which is a measure of the performance of the entire pipeline of our system, was 91.9%. Our system showed encouraging results for the conversion of free-text radiology reports into a structured format. The coverage of clinical information in the reports was 96.2% (6595/6853). Our 2-stage deep system can extract clinical information from chest and abdomen CT reports accurately and comprehensively.

Natural Language Processing for the Identification of Incidental Lung Nodules in Computed Tomography Reports: A Quality Control Tool.

To evaluate the diagnostic performance of a natural language processing (NLP) model in detecting incidental lung nodules (ILNs) in unstructured chest computed tomography (CT) reports. All unstructured consecutive reports of chest CT scans performed at a tertiary hospital between 2020 and 2021 were retrospectively reviewed (n = 21,542) to train the NLP tool. Internal validation was performed using reference readings by two radiologists of both CT scans and reports, using a different external cohort of 300 chest CT scans. Second, external validation was performed in a cohort of all random unstructured chest CT reports from 57 different hospitals conducted in May 2022. A review by the same thoracic radiologists was used as the gold standard. The sensitivity, specificity, and accuracy were calculated. Of 21,542 CT reports, 484 mentioned at least one ILN (mean age, 71 ± 17.6 [standard deviation] years; women, 52%) and were included in the training set. In the internal validation (n = 300), the NLP tool detected ILN with a sensitivity of 100.0% (95% CI, 97.6 to 100.0), a specificity of 95.9% (95% CI, 91.3 to 98.5), and an accuracy of 98.0% (95% CI, 95.7 to 99.3). In the external validation (n = 977), the NLP tool yielded a sensitivity of 98.4% (95% CI, 94.5 to 99.8), a specificity of 98.6% (95% CI, 97.5 to 99.3), and an accuracy of 98.6% (95% CI, 97.6 to 99.2). Twelve months after the initial reports, 8 (8.60%) patients had a final diagnosis of lung cancer, among which 2 (2.15%) would have been lost to follow-up without the NLP tool. NLP can be used to identify ILNs in unstructured reports with high accuracy, allowing a timely recall of patients and a potential diagnosis of early-stage lung cancer that might have been lost to follow-up.

Reporting breast density on chest computed tomography.

Women are encouraged to have a yearly mammogram and in addition to screening for breast cancer, the radiologist reports the patient's breast density. High breast density increases a woman's risk of developing breast cancer. The number of chest computed tomography (CT) scans performed each year is increasing. Chest CT scans for lung cancer screening in high-risk patients are the standard of care. Important additional findings can be identified on these exams including coronary artery calcifications, thyroid nodules, and breast density. Our previous research has shown that breast density can be reliably graded on chest CT and is comparable to mammographic grading. However, the inter-reader agreement was higher for chest CT. It is important that thoracic radiologists include the grading of breast density in their chest CT reports. According to mammography literature, this information has proven to be helpful for early detection of breast cancer. Federal legislation recommends notifying both providers and patients about breast density on mammography and so it follows that if we see the same information on chest CT, we should report it so that at the very least the clinician can encourage their patient to have a routine mammogram.

Prevalence and outcomes of interstitial lung abnormalities in a Canadian lung cancer screening trial

RATIONALE Interstitial lung abnormalities (ILA) are radiographic interstitial findings that are incidentally identified on chest imaging performed for other reasons. ILAs have been identified on chest computed tomography (CT) performed as part of lung cancer screening programs. OBJECTIVES This study characterizes the prevalence and outcomes associated with ILA in a Canadian lung cancer screening cohort. METHODS Patients were identified from the Alberta lung cancer screening trial, a 5-year cohort study. CT reports were screened for terms associated with ILA including specific radiologic features and CT patterns. These were further characterized as non-subpleural, subpleural nonfibrotic and subpleural fibrotic, consistent with published definitions. Retrospective chart review was conducted for those with ILA to define demographics, lung function, and longitudinal outcomes including subsequent investigations for interstitial lung disease (ILD), treatment, and survival. MEASUREMENTS AND MAIN RESULTS Of 806 patients in the lung cancer screening study, 30 (3.7%) were identified as having ILA, with two-thirds (67%) having subpleural fibrotic abnormalities. Half of patients were referred to a Respirologist and underwent pulmonary function testing. Over a median follow-up period of two years, none were diagnosed with an idiopathic interstitial pneumonia, or started on immunomodulatory or antifibrotic therapy. Three of 30 (10%) patients demonstrated disease progression over time, all of whom had subpleural fibrotic ILA on baseline chest CT. CONCLUSIONS The prevalence of ILA in this Canadian lung cancer screening cohort was 3.7%. These data should inform the development of standardized reporting and follow-up for ILA as lung cancer screening programs are implemented.

Enhancing the identification of rheumatoid arthritis-associated interstitial lung disease through text mining of chest computerized tomography reports

ObjectivesAlgorithms have been developed to identify rheumatoid arthritis-interstitial lung disease (RA-ILD) in administrative data with positive predictive values (PPVs) between 70 and 80%. We hypothesized that including ILD-related terms identified within chest computed tomography (CT) reports through text mining would improve the PPV of these algorithms in this cross-sectional study. MethodsWe identified a derivation cohort of possible RA-ILD cases (n = 114) using electronic health record data from a large academic medical center and performed medical record review to validate diagnoses (reference standard). ILD-related terms (e.g., ground glass, honeycomb) were identified in chest CT reports by natural language processing. Administrative algorithms including diagnostic and procedural codes as well as specialty were applied to the cohort both with and without the requirement for ILD-related terms from CT reports. We subsequently analyzed similar algorithms in an external validation cohort of 536 participants with RA. ResultsThe addition of ILD-related terms to RA-ILD administrative algorithms increased the PPV in both the derivation (improvement ranging from 3.6 to 11.7%) and validation cohorts (improvement 6.0 to 21.1%). This increase was greatest for less stringent algorithms. Administrative algorithms including ILD-related terms from CT reports exceeded a PPV of 90% (maximum 94.6% derivation cohort). Increases in PPV were accompanied by a decline in sensitivity (validation cohort -3.9 to -19.5%). ConclusionsThe addition of ILD-related terms identified by text mining from chest CT reports led to improvements in the PPV of RA-ILD algorithms. With high PPVs, use of these algorithms in large data sets could facilitate epidemiologic and comparative effectiveness research in RA-ILD.

How Natural Language Processing Can Aid With Pulmonary Oncology Tumor Node Metastasis Staging From Free-Text Radiology Reports: Algorithm Development and Validation

Natural language processing (NLP) is thought to be a promising solution to extract and store concepts from free text in a structured manner for data mining purposes. This is also true for radiology reports, which still consist mostly of free text. Accurate and complete reports are very important for clinical decision support, for instance, in oncological staging. As such, NLP can be a tool to structure the content of the radiology report, thereby increasing the report's value. This study describes the implementation and validation of an N-stage classifier for pulmonary oncology. It is based on free-text radiological chest computed tomography reports according to the tumor, node, and metastasis (TNM) classification, which has been added to the already existing T-stage classifier to create a combined TN-stage classifier. SpaCy, PyContextNLP, and regular expressions were used for proper information extraction, after additional rules were set to accurately extract N-stage. The overall TN-stage classifier accuracy scores were 0.84 and 0.85, respectively, for the training (N=95) and validation (N=97) sets. This is comparable to the outcomes of the T-stage classifier (0.87-0.92). This study shows that NLP has potential in classifying pulmonary oncology from free-text radiological reports according to the TNM classification system as both the T- and N-stages can be extracted with high accuracy.

Possibilities and limitations of using machine text-processing tools in Russian radiology reports

BACKGROUND: In radiology, important information can be found not only in medical images, but also in the accompanying text descriptions created by radiologists. Identification of study protocols containing certain data and extraction of these data can be useful primarily for clinical problems; however, given the large amount of such data, the development of machine analysis algorithms is necessary.&#x0D; AIM: To estimate the possibilities and limitations of using a tool for machine processing of radiology reports to search for pathological findings.&#x0D; MATERIALS AND METHODS: To create an algorithm for automatic analysis of radiology reports, use cases were selected that participated in the experiment on the use of innovative technologies in the computer vision for the analysis of medical images in 2020. Mammography, chest X-ray, chest computed tomography (CT), and LDCT, were among the use cases performed in Moscow. A dictionary of keywords has been compiled. After the automatic marking of the reports by the developed tool, the results were assessed by a radiologist. The number of protocols analyzed by the radiologist for training and validation of the algorithms was 977 for mammography, 4,804 for all chest X-ray scans, 4,074 for chest CT, and 398 for chest LDCT. For the final testing of the developed algorithms, test datasets of 1,032 studies for mammography, 544 for chest X-ray, 5,000 for CT of the chest, and 1,082 studies for the LDCT of the chest were additionally labeled.&#x0D; RESULTS: The best results were achieved in the search for viral pneumonia in chest CT reports (accuracy 0.996, sensitivity 0.998, and specificity 0.989) and breast cancer in mammography reports (accuracy 1.0, sensitivity 1.0, and specificity 1.0). When searching for signs of lung cancer by the algorithm, the metrics were as follows: accuracy 0.895, sensitivity 0.829, and specificity 0.936, when searching for pathological changes in the chest organs in radiography and fluorography protocols (accuracy 0.912, sensitivity 1.000, and specificity 0.844).&#x0D; CONCLUSIONS: Machine methods with high accuracy can be used to automatically classify the radiology reports of mammography and chest CT with viral pneumonia. The achieved accuracy is sufficient for successful application to automatically compare the conclusions of physicians and artificial intelligence models when searching for signs of lung cancer in chest CT and LDCT, pathological findings in chest X-ray.

Partial anomalous pulmonary venous return in adults: Insight into pulmonary hypertension

BackgroundPartial anomalous pulmonary venous return (PAPVR) is a rare condition that may lead to pulmonary arterial hypertension (PAH). We sought to determine the prevalence of PAPVR, the follow-up rate of incidentally discovered PAPVR, the repercussions of volume and pressure overload on the right ventricle (RV), and the prevalence and predictors of PAH. Methods and resultsFifty PAPVR patients aged ≥18 years were included. All underwent transthoracic echocardiography and multislice imaging. PAPVR prevalence was 0.2% (23/13,606 chest computed tomography reports). Of the 26 incidentally diagnosed patients, 14 (54%) did not have a cardiac follow-up. Seven (14%) patients had PAH, of whom one third (28%) had Eisenmenger syndrome. The left-to-right shunt ratio (Qp/Qs) was the only predictor of PAH. PAPVR-induced volume overload led to increased right chamber volumes (RV basal diameter 45±8 mm, RV mid-diameter 38±9 mm, RV/left ventricle ratio 1.2 ± 0.2, indexed RV end-diastolic area 14±4 cm2/m2, indexed RV end-systolic area 8.6 ± 3.2 cm2/m2) and increased RV functional parameters (tricuspid annular plane systolic excursion 27±5 mm, fractional area change [FAC] 42 ± 10%, global longitudinal strain [GLS] −22 ± 5). In contrast, PAH-induced pressure overload had no impact on right heart volumes, but on RV functional parameters, which were decreased compared to non-PAH patients (FAC 35 ± 13% vs. 43 ± 10%, p = 0.049; GLS -18 ± 5 vs. −23 ± 4%, p = 0.01). ConclusionsPAPVR prevalence was 0.2%. When incidentally diagnosed, its management was often neglected despite potentially serious consequences (14% PAH). Only the Qp/Qs ratio was predictive of PAH. PAPVR-induced volume overload was shown to increase right heart volumes, while PAH-induced pressure overload caused alterations of RV functional parameters.

Partial anomalous pulmonary venous return in adults. Insight into pulmonary hypertension

Partial anomalous pulmonary venous return (PAPVR) is a rare congenital heart disease (CHD) where pulmonary veins drain into pulmonary circulation leading to a left-to-right shunt. PAPVR can lead to serious consequences, ie pulmonary hypertension (PH). To date, predictors of PH are not well established. The aims of this study were (1) to determine the prevalence of PAPVR (2) to identify the impact of PAPVR on the right ventricle, (3) to determine the prevalence of PH in the PAPVR population and to identify its predictive factors. Patients aged ≥18 years with a non-operated PAPVR and patients operated after the age of 18 were enrolled. Patients with other potential causes of PH were excluded. PH was defined based on right heart catheterization (RHC). For patients without RHC, PH was considered when echocardiographic systolic transtricuspid peak velocity was ≥ 3.4 m/s. Fifty patients were included. Mean patient age was 50 ± 18 years with a 60% female predominance. PAPVR was mostly arising from the right lung (76% of cases) and was associated with CHD in 18 patients (36%). Most of them (83%) were sinus venosus atrial septal defect (ASD). PAPVR was diagnosed by chest computed tomography (CT) in 37 patients (74%). The reported prevalence of PAPVR was 0.2% (23/13606 chest CT reports). PH was identified in 7 patients of the cohort (14%) and 2 of them had Eisenmenger syndrome. All the echocardiographic parameters of RV dimensions were increased. PH group compared to group without PH had a decreased RV FAC (35 ± 13% vs 43 ± 10%, P = 0.049), a decreased RV endoGLS (−18 ± 5% vs −23 ± 4%, P = 0.01) and a significantly higher Qp/Qs ratio (2.2 [1.5;3.0] vs 1.1 [1.0;1.4], P = 0.013). Qp/Qs ratio was the only significant predictive factor for PH (OR = 4.87; P = 0.025), with an AUC of 0.87 ( P = 0.016) and the optimal threshold value was 1.95. Prevalence of PAPVR is 0.2% based on chest CT database. 14% of the patients with PAPVR developed PH and 30% of them had Einsenmenger syndrome. RV function parameters (FAC and endoGLS) were significantly decreased in PH group. Qp/Qs was the only significant predictive factor for PH.

French Imaging Database Against Coronavirus (FIDAC): A large COVID-19 multi-center chest CT database

PurposeDuring the first wave of the COVID-19 pandemic, the French Society of Radiology and the French College of Radiology, in partnership with NEHS Digital, have set up a system to collect chest computed tomography (CT) examinations with clinical, virological and radiological metadata, from patients clinically suspected of COVID-19 pneumonia. This allowed the constitution of an anonymized multicenter database, named FIDAC (French Imaging Database Against Coronavirus). The aim of this report was to describe the content of this public database. Materials and methodsTwenty-two French radiology centers participated to the data collection. The data collected were chest CT examinations in DICOM format associated with the following metadata: patient age and sex, originating facility identifier, originating facility region, time from symptom onset to CT examination, indication for CT examination, reverse transcription-polymerase chain reaction (RT-PCR) results and normalized CT report performed by a senior radiologist. All the data were anonymized and sent through a NEHS Digital system to a centralized data center. ResultsA total of 5944 patients were included from the 22 centers aggregated into 8 regions with a mean number of patients of 743 ± 603.3 [SD] per region (range: 102–1577 patients). Reasons for CT examination and normalized CT reports were provided for all patients. RT-PCR results were provided in 5574 patients (93.77%) with a positive result of RT-PCR in 44.6% of patients. ConclusionThe FIDAC project allowed the creation of a large database of chest CT images and metadata available, under conditions, in open access through the CERF-SFR website.

Incidental findings in chest computed tomography of patients with thoracic trauma: what we need to know.

This study emphasizes the importance of appropriately evaluating incidental findings (IF) detected in chest computed tomography (CCTs) of patients with trauma and provides related guidance and information. All patients aged ≥ 18 years who were admitted to the emergency department and underwent CCT for thoracic trauma were retrospectively screened, and 567 patients were included in the study. The findings were divided into five groups: those requiring immediate intervention, clinical evaluation, additional examination, and control examination after follow-up and those not requiring follow-up. Additionally, to evaluate the emergency reporting of these pathologies, the CCT reports were evaluated to reveal the IF rate. The mean age of the 567 patients was 47.96 ± 19.38 years (18-102 years); of which, 372 (65.6%) were male and 195 (34.4%) were female. At least one IF was observed in 261 (46%) patients. The lungs of 81 (13.6%) patients exhibited the highest rate of IFs. No difference was observed between males and females in terms of the presence of at least one IF (p = 0.144). The mean age of the patients with IF was significantly higher than that of those without IF (p < 0.001). Of the 294 patients whose CCT was presented in an emergency report by a radiologist, 142 (48.6%) had at least one IF; however, no IF was mentioned in the reports of 77 patients (54.5%). IFs should be properly reported according to their severity and included in patient records, consequently facilitating early treatment or additional examinations and eliminating the need for performing repeated examinations for those not requiring follow-up.

Improvement in Stage of Lung Cancer Diagnosis With Incidental Pulmonary Nodules Followed With a Patient Tracking System and Computerized Registry

IntroductionGiven that an incidental pulmonary nodule (IPN) on chest computed tomography (CT) may represent nascent lung cancer, timely follow-up imaging is critical to assess nodule growth and the need for tissue sampling. We previously reported our institution’s systematic process to identify and track patients with an IPN associated with improved CT on follow-up. We hypothesized that this improvement may have led to a higher frequency of early-stage lung cancer. To evaluate this, we performed a study to determine whether cases of early-stage lung cancer were more likely to have had our tracking system applied to suspicious findings. MethodsAn observational study was performed by identifying cases of lung cancer that were detected as IPNs on chest CT scans performed at our institution, from 2006 to 2016. A total of 314 cases were dichotomized into early-stage (stage 1) or late-stage (stages II to IV) disease. A multivariant regression analysis with modeling was used to determine factors associated with a diagnosis of early-stage disease. Factors included the use of the tracking system and nodule registry. ResultsThe following factors were independently associated with early-stage lung cancer: index nodule diameter, (OR = 0.971, confidence interval [CI]: 0.948–0.995], p = 0.016), adenocarcinoma histology (OR = 2.930 [CI: 1.695–5.064], p = 0.0001) and use of tracker phrases on CT reports (OR = 1.939 [CI: 1.126–3.339], p = 0.016). ConclusionsThe application of a patient tracking system and computerized lung nodule registry lead to an increased frequency in the diagnosis of stage 1 NSCLC from IPNs. This is a meaningful outcome for patients and should be adapted for IPN management.

Automatic Extraction of Lung Cancer Staging Information From Computed Tomography Reports: Deep Learning Approach.

BackgroundLung cancer is the leading cause of cancer deaths worldwide. Clinical staging of lung cancer plays a crucial role in making treatment decisions and evaluating prognosis. However, in clinical practice, approximately one-half of the clinical stages of lung cancer patients are inconsistent with their pathological stages. As one of the most important diagnostic modalities for staging, chest computed tomography (CT) provides a wealth of information about cancer staging, but the free-text nature of the CT reports obstructs their computerization.ObjectiveWe aimed to automatically extract the staging-related information from CT reports to support accurate clinical staging of lung cancer.MethodsIn this study, we developed an information extraction (IE) system to extract the staging-related information from CT reports. The system consisted of the following three parts: named entity recognition (NER), relation classification (RC), and postprocessing (PP). We first summarized 22 questions about lung cancer staging based on the TNM staging guideline. Next, three state-of-the-art NER algorithms were implemented to recognize the entities of interest. Next, we designed a novel RC method using the relation sign constraint (RSC) to classify the relations between entities. Finally, a rule-based PP module was established to obtain the formatted answers using the results of NER and RC.ResultsWe evaluated the developed IE system on a clinical data set containing 392 chest CT reports collected from the Department of Thoracic Surgery II in the Peking University Cancer Hospital. The experimental results showed that the bidirectional encoder representation from transformers (BERT) model outperformed the iterated dilated convolutional neural networks-conditional random field (ID-CNN-CRF) and bidirectional long short-term memory networks-conditional random field (Bi-LSTM-CRF) for NER tasks with macro-F1 scores of 80.97% and 90.06% under the exact and inexact matching schemes, respectively. For the RC task, the proposed RSC showed better performance than the baseline methods. Further, the BERT-RSC model achieved the best performance with a macro-F1 score of 97.13% and a micro-F1 score of 98.37%. Moreover, the rule-based PP module could correctly obtain the formatted results using the extractions of NER and RC, achieving a macro-F1 score of 94.57% and a micro-F1 score of 96.74% for all the 22 questions.ConclusionsWe conclude that the developed IE system can effectively and accurately extract information about lung cancer staging from CT reports. Experimental results show that the extracted results have significant potential for further use in stage verification and prediction to facilitate accurate clinical staging.

An Intervention to Tag Findings Suspicious for Lung Cancer on Chest Computed Tomography Has Good Sensitivity and Number Needed to Diagnose.

In 2015, Kaiser Permanente Northern California implemented an intervention to improve follow-up for pulmonary findings on diagnostic chest computed tomography (CT). The intervention includes tagging CT reports with the prefix "#PUL" followed by a character (0-6 or X) to track specific findings. #PUL5, indicating "suspicious for malignancy," triggers automatic referral for multidisciplinary care review. Among patients who obtained an index chest CT exam from August 2015 to July 2017 without an exam in the previous 2 years, we computed the frequency of lung cancer diagnosis within 120 days of CT in relation to each #PUL tag. For #PUL5, we computed sensitivity, specificity, positive and negative predictive values, and number needed to diagnose. We also performed a chart review to assess why some patients diagnosed with lung cancer were not tagged #PUL5. Of the 39,409 patients with a tagged CT report, 1105 (2.8%) had a new primary lung cancer diagnosis within 120 days. Among the 2255 patients tagged #PUL5, 821 were diagnosed with lung cancer, with a sensitivity of 74% (95% confidence interval, 72%-77%). The positive predictive value was 36% (35%-38%), number needed to diagnosis was 2.7 (2.6-2.9), and specificity and negative predictive values were > 95%. Chart review identified opportunities to improve system defaults and clarify concepts. The intervention performed well but needed improvement. Automating CT reports was simple and generalizable, and enabled reduction of care gaps and system improvement.

Extracting clinical terms from radiology reports with deep learning

Extracting clinical terms from free-text format radiology reports is a first important step toward their secondary use. However, there is no general consensus on the kind of terms to be extracted. In this paper, we propose an information model comprising three types of clinical entities: observations, clinical findings, and modifiers. Furthermore, to determine its applicability for in-house radiology reports, we extracted clinical terms with state-of-the-art deep learning models and compared the results. We trained and evaluated models using 540 in-house chest computed tomography (CT) reports annotated by multiple medical experts. Two deep learning models were compared, and the effect of pre-training was explored. To investigate the generalizability of the model, we evaluated the use of other institutional chest CT reports. The micro F1-score of our best performance model using in-house and external datasets were 95.36% and 94.62%, respectively. Our results indicated that entities defined in our information model were suitable for extracting clinical terms from radiology reports, and the model was sufficiently generalizable to be used with dataset from other institutions.