Natural language processing pipeline to extract prostate cancer-related information from clinical notes.

OBJECTIVES: To develop an automated pipeline for extracting prostate cancer-related information from clinical notes. MATERIALS AND METHODS: This retrospective study included 23,225 patients who underwent prostate MRI between 2017 and 2022. Cancer risk factors (family history of cancer and digital rectal exam findings), pre-MRI prostate pathology, and treatment history of prostate cancer were extracted from free-text clinical notes in English as binary or multi-class classification tasks. Any sentence containing pre-defined keywords was extracted from clinical notes within one year before the MRI. After manually creating sentence-level datasets with ground truth, Bidirectional Encoder Representations from Transformers (BERT)-based sentence-level models were fine-tuned using the extracted sentence as input and the category as output. The patient-level output was determined by compilation of multiple sentence-level outputs using tree-based models. Sentence-level classification performance was evaluated using the area under the receiver operating characteristic curve (AUC) on 15% of the sentence-level dataset (sentence-level test set). The patient-level classification performance was evaluated on the patient-level test set created by radiologists by reviewing the clinical notes of 603 patients. Accuracy and sensitivity were compared between the pipeline and radiologists. RESULTS: Sentence-level AUCs were ≥ 0.94. The pipeline showed higher patient-level sensitivity for extracting cancer risk factors (e.g., family history of prostate cancer, 96.5% vs. 77.9%, p < 0.001), but lower accuracy in classifying pre-MRI prostate pathology (92.5% vs. 95.9%, p = 0.002) and treatment history of prostate cancer (95.5% vs. 97.7%, p = 0.03) than radiologists, respectively. CONCLUSION: The proposed pipeline showed promising performance, especially for extracting cancer risk factors from patient's clinical notes. CLINICAL RELEVANCE STATEMENT: The natural language processing pipeline showed a higher sensitivity for extracting prostate cancer risk factors than radiologists and may help efficiently gather relevant text information when interpreting prostate MRI. KEY POINTS: When interpreting prostate MRI, it is necessary to extract prostate cancer-related information from clinical notes. This pipeline extracted the presence of prostate cancer risk factors with higher sensitivity than radiologists. Natural language processing may help radiologists efficiently gather relevant prostate cancer-related text information.

MRI in Adult Patients with Active and Inactive Implanted MR-conditional, MR-nonconditional, and Other Devices.

Active implanted medical devices (AIMDs) enable therapy and patient monitoring by way of electrical activity and typically have a battery and electrical leads. The most common types of AIMDs include cardiac implantable electronic devices (CIEDs), spinal cord stimulators, deep brain stimulators, bone growth or fusion stimulators, other neurostimulators, and drug infusion pumps. As more patients with AIMDs undergo MRI, it is important to consider the safety of patients who have these implanted devices during MRI. The authors review the physics concepts related to MRI safety, such as peak spatial gradient magnetic field, specific absorption rate, root mean square value of the effective magnetic component of the transmitted RF pulse, and gradient slew rate, as well as the parameters necessary to remain within safety limits. The roles of MRI safety personnel, as set forth by the International Society of Magnetic Resonance in Medicine, are emphasized. In addition, the relevant information provided in vendor manuals is reviewed, with a focus on how to obtain relevant up-to-date information. The radiologist should be able to modify protocols to meet safety requirements, address possible alternatives to MRI, and weigh the potential benefits of MRI against the potential risks. A few more advanced topics, such as fractured or abandoned device leads and patients with multiple implanted medical devices, also are addressed. Recommended workflows for MRI in patients with implanted medical devices are outlined. It is important to implement an algorithmic MRI safety process, including a review of the MRI safety information; patient screening; optimal imaging; and monitoring patients before, during, and after the examination. ©RSNA, 2024 Test Your Knowledge questions for this article are available in the supplemental material. See the invited commentary by Shetty et al in this issue.

Diagnosis of Usual Interstitial Pneumonitis in the Absence of Honeycombing: Evaluation of Specific CT Criteria With Clinical Follow-Up in 38 Patients.

OBJECTIVE: We sought to evaluate specific CT criteria for the diagnosis of usual interstitial pneumonitis (UIP) in the absence of honeycombing. These criteria included peripheral reticulation and lobular distortion; some upper lobe involvement, but a lower zone predominance; a heterogeneous appearance with areas of normal lung, minimal reticulation, and substantial distortion alternating throughout the study and often on an individual image; a nonsegmental distribution; and traction bronchiectasis. MATERIALS AND METHODS: We searched reports of CT studies performed between January 1, 2009, and January 1, 2012, to identify patients for whom UIP was a likely or probable diagnosis and reviewed the CT study for each case (n = 106). There were 38 patients who met all CT criteria and who also had a clinical diagnosis of idiopathic UIP (also known as idiopathic pulmonary fibrosis [IPF]) and follow-up of at least 6 months, as determined from the electronic medical record. We reviewed prior and subsequent CT examinations in this cohort. RESULTS: The median age of our patients was 80 years, and the duration of clinical follow-up was 6-104 months (mean, 38 months; median, 37 months). For all patients, a pulmonary medicine physician made a working diagnosis of IPF. Fifteen patients died from pulmonary complications, and 16 of the surviving patients had clinical or functional progression of disease. There were no instances in which the initial diagnosis was revised or reversed. CONCLUSION: Strict application of specific CT criteria may allow a specific diagnosis of UIP in the proper clinical setting in the absence of honeycombing.

Automated detection of z-axis coverage with abdomen-pelvis computed tomography examinations.

Excessive cephalocaudal anatomic (Z-axis) coverage can lead to unnecessary radiation exposure to a patient. In this study, an automated computing model was developed for identifying instances of potentially excessive Z-axis coverage with abdomen-pelvis examinations. Eight patient and imaging attributes including patient gender, age, height, weight, volume CT dose index (CTDIvol), dose length product (DLP), maximum abdomen width, and maximum abdomen thickness were used to build a feedforward neural network model to predict a target Z-axis coverage whether it is an excessive or non-excessive Z-axis coverage scans. 264 CT abdomen-pelvis exams were used to develop the model which is validated using 10-fold cross validation. The result showed that 244 out of 264 exams (92.4%) correctly predicted Z-axis excessive coverage. The promising results indicate that this tool has the potential to be used for CT exams of the chest and colon, urography, and other site-specified CT studies having defined limited length.

Reducing body CT radiation dose: beyond just changing the numbers.

OBJECTIVE: CT dose reduction has become a top priority for many radiology practices as a result of federal and state initiatives and public concern. Implementing this in practice, however, is difficult because of the variability between practices, CT scanners, radiologist preferences, and institutional capacity. CONCLUSION: This article will discuss strategies for successful CT dose reduction instituted in multivendor practices.

An automated DICOM database capable of arbitrary data mining (including radiation dose indicators) for quality monitoring.

The U.S. National Press has brought to full public discussion concerns regarding the use of medical radiation, specifically x-ray computed tomography (CT), in diagnosis. A need exists for developing methods whereby assurance is given that all diagnostic medical radiation use is properly prescribed, and all patients' radiation exposure is monitored. The "DICOM Index Tracker©" (DIT) transparently captures desired digital imaging and communications in medicine (DICOM) tags from CT, nuclear imaging equipment, and other DICOM devices across an enterprise. Its initial use is recording, monitoring, and providing automatic alerts to medical professionals of excursions beyond internally determined trigger action levels of radiation. A flexible knowledge base, aware of equipment in use, enables automatic alerts to system administrators of newly identified equipment models or software versions so that DIT can be adapted to the new equipment or software. A dosimetry module accepts mammography breast organ dose, skin air kerma values from XA modalities, exposure indices from computed radiography, etc. upon receipt. The American Association of Physicists in Medicine recommended a methodology for effective dose calculations which are performed with CT units having DICOM structured dose reports. Web interface reporting is provided for accessing the database in real-time. DIT is DICOM-compliant and, thus, is standardized for international comparisons. Automatic alerts currently in use include: email, cell phone text message, and internal pager text messaging. This system extends the utility of DICOM for standardizing the capturing and computing of radiation dose as well as other quality measures.

Radiation dose to the fetus for pregnant patients undergoing multidetector CT imaging: Monte Carlo simulations estimating fetal dose for a range of gestational age and patient size.

PURPOSE: To use Monte Carlo simulations of a current-technology multidetector computed tomographic (CT) scanner to investigate fetal radiation dose resulting from an abdominal and pelvic examination for a range of actual patient anatomies that include variation in gestational age and maternal size. MATERIALS AND METHODS: Institutional review board approval was obtained for this HIPAA-compliant retrospective study. Twenty-four models of maternal and fetal anatomy were created from image data from pregnant patients who had previously undergone clinically indicated CT examination. Gestational age ranged from less than 5 weeks to 36 weeks. Simulated helical scans of the abdominal and pelvic region were performed, and a normalized dose (in milligrays per 100 mAs) was calculated for each fetus. Stepwise multiple linear regression was performed to analyze the correlation of dose with gestational age and anatomic measurements of maternal size and fetal location. Results were compared with several existing fetal dose estimation methods. RESULTS: Normalized fetal dose estimates from the Monte Carlo simulations ranged from 7.3 to 14.3 mGy/100 mAs, with an average of 10.8 mGy/100 mAs. Previous methods yielded values of 10-14 mGy/100 mAs. The correlation between gestational age and fetal dose was not significant (P = .543). Normalized fetal dose decreased linearly with increasing patient perimeter (R(2) = 0.681, P < .001), and a two-factor model with patient perimeter and fetal depth demonstrated a strong correlation with fetal dose (R(2) = 0.799, P < .002). CONCLUSION: A method for the estimation of fetal dose from models of actual patient anatomy that represented a range of gestational age and patient size was developed. Fetal dose correlated with maternal perimeter and varied more than previously recognized. This correlation improves when maternal size and fetal depth are combined.

Three-dimensional virtual dissection at CT colonography: unraveling the colon to search for lesions.

Computed tomographic (CT) colonography is a promising noninvasive examination for colorectal cancer screening; however, the optimal interpretation strategy remains undecided. Virtual dissection is an innovative technique whereby the three-dimensional (3D) model of the colon is virtually unrolled, sliced open, and displayed as a flat 3D rendering of the mucosal surface, similar to a gross pathologic specimen. This technique has the potential to reduce evaluation time by providing a more rapid 3D image assessment than is possible with an antegrade and retrograde 3D endoluminal fly-through. It may also ultimately improve accuracy by reducing blind spots present with 3D endoluminal displays and by reducing reader fatigue. A disadvantage of virtual dissection is the potential for distortion of colonic lesions and normal anatomy. To avoid potential pitfalls in image interpretation, the radiologist must be familiar with the unique appearance of the normal colon anatomy and of various pathologic findings when using virtual dissection with two-dimensional axial and 3D endoluminal CT colonographic image data sets.

Peripheral (deep) but not periventricular MRI white matter hyperintensities are increased in clinical vascular dementia compared to Alzheimer's disease.

BACKGROUND AND PURPOSE: Vascular dementia (VAD) is a complex diagnosis at times difficult to distinguish from Alzheimer's disease (AD). MRI scans often show white matter hyperintensities (WMH) in both conditions. WMH increase with age, and both VAD and AD are associated with aging, thus presenting an attribution conundrum. In this study, we sought to show whether the amount of WMH in deep white matter (dWMH), versus periventricular white matter (PVH), would aid in the distinction between VAD and AD, independent of age. METHODS: Blinded semiquantitative ratings of WMH validated by objective quantitation of WMH volume from standardized MRI image acquisitions. PVH and dWMH were rated separately and independently by two different examiners using the Scheltens scale. Receiver operator characteristic (ROC) curves were generated using logistic regression to assess classification of VAD (13 patients) versus AD (129 patients). Clinical diagnoses were made in a specialty memory disorders clinic. RESULTS: Using PVH rating alone, overall classification (area under the ROC curve, AUC) was 75%, due only to the difference in age between VAD and AD patients in our study and not PVH. In contrast, dWMH rating produced 86% classification accuracy with no independent contribution from age. A global Longstreth rating that combines dWMH and PVH gave an 88% AUC. CONCLUSIONS: Increased dWMH indicate a higher likelihood of VAD versus AD. Assessment of dWMH on MRI scans using Scheltens and Longstreth scales may aid the clinician in distinguishing the two conditions.

Value of multislice computed tomography angiography of the thorax in preparation for catheter ablation for the treatment of atrial fibrillation: the impact of unexpected cardiac and extracardiac findings on patient care.

OBJECTIVE: In patients referred for catheter ablation for the treatment of atrial fibrillation, multislice computed tomography angiography of the thorax is routinely performed to assess pulmonary vein anatomy. We sought to investigate the incidence of unexpected cardiac and extracardiac findings in this select patient population and to establish how these findings influence subsequent patient care. METHODS: Ninety-five patients (mean age 62+/-10 years, 35% female) referred to our institution for ablation therapy for atrial fibrillation between July 2003 and October 2007 underwent multislice computed tomography angiography of the thorax. Radiologists interpreted all images. Need for additional testing, consultation and eventual diagnosis were assessed by electronic record review. RESULTS: A total of 83 (5 cardiac, 78 extracardiac) unexpected findings were observed in 50/95 (53%) of patients. The findings prompted 23 additional tests (5 cardiac, 18 noncardiac) in 15/95 (16%) of patients and 8 subsequent referrals in 7/95 (7%) patients. In 6 patients the findings significantly altered future patient care and resulted in postponement of ablation therapy in 4 patients. In 2 patients, extracardiac findings (pulmonary emboli and adenocarcinoma of the lung) were of potentially life-saving consequence. CONCLUSIONS: In patients undergoing multislice computed tomography angiography of the thorax in anticipation of planned catheter ablation therapy for the treatment of atrial fibrillation, unexpected findings are common and of potentially significant value. In comparison, there is a higher prevalence of unexpected extracardiac, rather than cardiac findings. Further investigation of these findings may lead to postponement of ablation therapy, but may also be of potentially lifesaving consequence.

Extensive cardiac metastases secondary to uterine leiomyosarcoma.

Uterine leiomyosarcoma metastatic to multiple cardiac chambers occurs infrequently. Over the past 4 decades, there have been only 11 reported cases of cardiac metastasis secondary to uterine leiomyosarcoma. Metastatic cardiac tumors are commonly diagnosed postmortem, although it may be possible to diagnose them during life. The authors report a rare case of a 48-year-old woman with uterine leiomyosarcoma who was diagnosed with extensive cardiac metastases (in the left atrium, right and left ventricles, pulmonary artery, and pulmonary vein). Although extensive cardiac metastases are suggestive of imminent death, aggressive treatment resulted in survival for >1 year for this patient, which underscores the importance of treatment in select patients.