RESUMEN

Recommendations are a key component of radiology reports. Automatic extraction of recommendations would facilitate tasks such as recommendation tracking, quality improvement, and large-scale descriptive studies. Existing report-parsing systems are frequently limited to recommendations for follow-up imaging studies, operate at the sentence or document level rather than the individual recommendation level, and do not extract important contextualizing information. We present a neural network architecture capable of extracting fully contextualized recommendations from any type of radiology report. We identified six major "questions" necessary to capture the majority of context associated with a recommendation: recommendation, time period, reason, conditionality, strength, and negation. We developed a unified task representation by allowing questions to refer to answers to other questions. Our representation allows for a single system to perform named entity recognition (NER) and classification tasks. We annotated 2272 radiology reports from all specialties, imaging modalities, and multiple hospitals across our institution. We evaluated the performance of a long short-term memory (LSTM) architecture on the six-question task. The single-task LSTM model achieves a token-level performance of 89.2% at recommendation extraction, and token-level performances between 85 and 95% F1 on extracting modifying features. Our model extracts all types of recommendations, including follow-up imaging, tissue biopsies, and clinical correlation, and can operate in real time. It is feasible to extract complete contextualized recommendations of all types from arbitrary radiology reports. The approach is likely generalizable to other clinical entities referenced in radiology reports, such as radiologic findings or diagnoses.

Asunto(s)

RESUMEN

While radiologists regularly issue follow-up recommendations, our preliminary research has shown that anywhere from 35 to 50% of patients who receive follow-up recommendations for findings of possible cancer on abdominopelvic imaging do not return for follow-up. As such, they remain at risk for adverse outcomes related to missed or delayed cancer diagnosis. In this study, we develop an algorithm to automatically detect free text radiology reports that have a follow-up recommendation using natural language processing (NLP) techniques and machine learning models. The data set used in this study consists of 6000 free text reports from the author's institution. NLP techniques are used to engineer 1500 features, which include the most informative unigrams, bigrams, and trigrams in the training corpus after performing tokenization and Porter stemming. On this data set, we train naive Bayes, decision tree, and maximum entropy models. The decision tree model, with an F1 score of 0.458 and accuracy of 0.862, outperforms both the naive Bayes (F1 score of 0.381) and maximum entropy (F1 score of 0.387) models. The models were analyzed to determine predictive features, with term frequency of n-grams such as "renal neoplasm" and "evalu with enhanc" being most predictive of a follow-up recommendation. Key to maximizing performance was feature engineering that extracts predictive information and appropriate selection of machine learning algorithms based on the feature set.

Asunto(s)

RESUMEN

Unstructured and semi-structured radiology reports represent an underutilized trove of information for machine learning (ML)-based clinical informatics applications, including abnormality tracking systems, research cohort identification, point-of-care summarization, semi-automated report writing, and as a source of weak data labels for training image processing systems. Clinical ML systems must be interpretable to ensure user trust. To create interpretable models applicable to all of these tasks, we can build general-purpose systems which extract all relevant human-level assertions or "facts" documented in reports; identifying these facts is an information extraction (IE) task. Previous IE work in radiology has focused on a limited set of information, and extracts isolated entities (i.e., single words such as "lesion" or "cyst") rather than complete facts, which require the linking of multiple entities and modifiers. Here, we develop a prototype system to extract all useful information in abdominopelvic radiology reports (findings, recommendations, clinical history, procedures, imaging indications and limitations, etc.), in the form of complete, contextualized facts. We construct an information schema to capture the bulk of information in reports, develop real-time ML models to extract this information, and demonstrate the feasibility and performance of the system.

Asunto(s)

RESUMEN

OBJECTIVE: The effect of demographics and societal determinants on imaging follow-up rates is not clear. The purpose of this study was to compare characteristics of patients with imaging findings representing possible cancer who undergo follow-up imaging versus those who do not to better understand factors that contribute to follow-up completion. MATERIALS AND METHODS: The records of 1588 patients with indeterminate abdominal imaging findings consecutively registered between July 1, 2013, and March 20, 2014, were reviewed. Several patient characteristics, including distance between patients' home zip codes and the flagship hospital of the health system were compared between the groups who did and did not undergo follow-up imaging. Subgroup analyses based on the location of the index examination were also performed. RESULTS: Among the 1513 (36.62%) included patients, 554 did not undergo follow-up abdominal imaging within 1 year of the index examination. The same was true of 270 of 938 (28.78%) outpatients and 168 of 279 (60.21%) emergency department patients. Eighty-nine of 959 (9.28%) patients who underwent follow-up imaging were younger than 40 years, compared with 76 of 554 (13.72%) patients who did not undergo follow-up imaging (p = 0.005). Fifty-four of 959 (5.63%) patients who underwent follow-up imaging were older than 80 years, compared with 70 of 554 (12.64%) patients who did not undergo follow-up imaging (p < 0.001). More white patients (587 of 959 vs 301 of 554, p = 0.007) and fewer black patients (204 of 554 versus 270 of 959, p < 0.001) were found in the follow-up imaging group. Greater distance from the flagship hospital correlated with less follow-up in the outpatient subgroup only (p = 0.03). CONCLUSION: Emergency department patients and patients at the extremes of age are less likely to complete follow-up imaging. Insurance status and race and ethnicity may affect follow-up completion rates. The relationship between distance to hospital and follow-up completion requires further investigation.

Asunto(s)

RESUMEN

PURPOSE: To evaluate the relationship between patient location at time of imaging and completion of relevant imaging follow-up for findings with indeterminate malignant potential. METHODS: We used a mandatory hospital-wide standardized assessment categorization system to analyze all ultrasound, CT, and MRI examinations performed over a 7-month period. Multivariate logistic regression, adjusted for imaging modality, characteristics of patients, ordering clinicians, and interpreting radiologists, was used to evaluate the relationship between patient location (outpatient, inpatient, or emergency department) at the time of index examination and completion of relevant outpatient imaging follow-up. RESULTS: Relevant follow-up occurred in 49% of index examinations, with a greater percentage among those performed in the outpatient setting compared with those performed in the inpatient or emergency department settings (62% versus 18% versus 17%, respectively). Compared with examinations obtained in the outpatient setting, examinations performed in the emergency department (adjusted odds ratio [aOR] 0.07; 95% confidence interval [CI], 0.03-0.19) and inpatient (aOR 0.14; 95% CI, 0.09-0.23) settings were less likely to be followed up. Black patients and those residing in lower-income neighborhoods were also less likely to receive relevant follow-up. Few lesions progressed to more suspicious lesions (4.6%). CONCLUSIONS: Patient location at time of imaging is associated with the likelihood of completing relevant follow-up imaging for lesions with indeterminate malignant potential. Future work should evaluate health system-level care processes related to care setting, as well as their effects on appropriate follow-up imaging. Doing so would support efforts to improve appropriate follow-up imaging and reduce health care disparities.

Asunto(s)

RESUMEN

PURPOSE: To evaluate the performance of machine learning algorithms on organ-level classification of semistructured pathology reports, to incorporate surgical pathology monitoring into an automated imaging recommendation follow-up engine. MATERIALS AND METHODS: This retrospective study included 2013 pathology reports from patients who underwent abdominal imaging at a large tertiary care center between 2012 and 2018. The reports were labeled by two annotators as relevant to four abdominal organs: liver, kidneys, pancreas and/or adrenal glands, or none. Automated classification methods were compared: simple string matching, random forests, extreme gradient boosting, support vector machines, and two neural network architectures-convolutional neural networks and long short-term memory networks. Three methods from the literature were used to provide interpretability and qualitative validation of the learned network features. RESULTS: The neural networks performed well on the four-organ classification task (F1 score: 96.3% for convolutional neural network and 96.7% for long short-term memory vs 89.9% for support vector machines, 93.9% for extreme gradient boosting, 82.8% for random forests, and 75.2% for simple string matching). Multiple methods were used to visualize the decision-making process of the network, verifying that the networks used similar heuristics to a human annotator. The neural networks were able to classify, with a high degree of accuracy, pathology reports written in unseen formats, suggesting the networks had learned a generalizable encoding of the salient features. CONCLUSION: Neural network-based approaches achieve high performance on organ-level pathology report classification, suggesting that it is feasible to use them within automated tracking systems.© RSNA, 2019Supplemental material is available for this article.See also the commentary by Liu in this issue.

RESUMEN

PURPOSE: The purpose of this study was to evaluate the relationship between final outcome of lesions indeterminate for malignancy on ultrasound (US) and patient and imaging characteristics. METHODS: We identified all patients with indeterminate liver lesions on US between 9/1/2013 and 12/31/2014 using institutional codes based on radiologist opinion. Miscoded lesions (n = 30) and patients with no imaging, pathology, or clinical follow-up at our health system (n = 6) were excluded. Final diagnostic category of malignant, benign, pseudolesion, or indeterminate was assigned using imaging, pathology, and clinical follow-up. Differences in diagnostic categories were compared by patient (age, gender, race, known malignancy. or liver disease) and imaging characteristics (lesion size, echogenicity. and number). Independent likelihood of a benign final diagnostic category was adjusted for significant variables on univariate analysis. RESULTS: Indeterminate liver lesions on US were found in 153/6813 patients (2%). Final diagnostic categories were malignant (11/153, 7%), benign (94/153, 61%), pseudolesion (42/153, 27%). and indeterminate (6/153, 4%). Nearly one-third of hypoechoic masses in patients with known malignancy or liver disease (i.e., high-risk status) ≥ 46 years of age were malignant (9/28, 32%). On multivariate analysis, patients of age ≥ 61 years and high-risk status were associated with decreased likelihood of benign diagnostic category (OR .19 (95% CI .07-.51) and OR .40 (95% CI .18-.88), p values .001 and .022, respectively). CONCLUSIONS: 2% of patients undergoing abdominal US have sonographically indeterminate liver lesions, of which 7% are malignant. Older, high-risk patients with hypoechoic lesions should receive short-term follow-up as one-third will have malignant lesions. Younger, low-risk patients should receive conservative follow-up, regardless of US imaging features.

Asunto(s)

RESUMEN

PURPOSE: The aims of this study were to examine the association between radiologist-initiated verbal communication for abdominal imaging findings indeterminate for malignancy and receipt of relevant outpatient follow-up imaging and to evaluate the proportion of findings that progress from indeterminate to suspicious for malignancy. METHODS: Using a mandatory standardized assessment coding system, 727 eligible outpatient abdominal CT, MRI, and ultrasound studies performed between July 1, 2013, and January 31, 2014, were identified. Multivariate logistic regression was used to evaluate the association between radiologist-initiated verbal communication and receipt of relevant outpatient imaging follow-up, adjusted for patient, ordering physician, interpreting radiologist, and imaging variables. RESULTS: Sixty-two percent of patients (451 of 727) with imaging findings indeterminate for malignancy received relevant outpatient imaging follow-up. Radiologist-initiated verbal communication occurred in 4.8% of cases (35 of 727). Radiologist-initiated verbal communication (odds ratio [OR], 2.45; 95% confidence interval [CI], 1.02-5.87) and existing cancer diagnosis (OR, 3.00; 95% CI, 2.11-4.26) were associated with a higher likelihood of receiving relevant imaging follow-up. Compared with those whose imaging studies were ordered by clinicians in a medical specialty, patients with studies ordered by clinicians in surgical (OR, 0.70; 95% CI, 0.49-0.99) or other (OR, 0.44; 95% CI, 20.24-0.83) specialties were less likely to receive relevant imaging follow-up. Progression of findings from indeterminate to suspicious for malignancy occurred in 5.4% of cases and was not associated with radiologist-initiated verbal communication. CONCLUSIONS: Radiologist-initiated verbal communication increases the likelihood that patients receive outpatient imaging follow-up for abdominal findings indeterminate for malignancy. Progression of these findings is relatively infrequent.

Asunto(s)

Asunto(s)

Asunto(s)

RESUMEN

An automated radiology recommendation-tracking engine for incidental focal masses in the liver, pancreas, kidneys, and adrenal glands was launched within our institution in July 2013. For 2 years, the majority of CT, MR, and US examination reports generated within our health system were mined by the engine. However, the need to expand the system beyond the initial four organs was soon identified. In July 2015, the second phase of the system was implemented and expanded to include additional anatomic structures in the abdomen and pelvis, as well as to provide non-radiology and non-imaging options for follow-up. The most frequent organs with incidental findings, outside of the original four, included the ovaries and the endometrium, which also correlated to the most frequently ordered imaging follow-up study of pelvic ultrasound and non-imaging follow-up study of endometrial biopsies, respectively. The second phase expansion has demonstrated new venues for augmenting and improving radiologist roles in optimal communication and management of incidental findings.

Asunto(s)

Asunto(s)

Asunto(s)