Oncology Scan Research Articles

Enhancing Radiological Reporting in Head and Neck Cancer: Converting Free-Text CT Scan Reports to Structured Reports Using Large Language Models

Objective The aim of this study was to assess efficacy of large language models (LLMs) for converting free-text computed tomography (CT) scan reports of head and neck cancer (HNCa) patients into a structured format using a predefined template. Materials and Methods A retrospective study was conducted using 150 CT reports of HNCa patients. A comprehensive structured reporting template for HNCa CT scans was developed, and the Generative Pre-trained Transformer 4 (GPT-4) was initially used to convert 50 CT reports into a structured format using this template. The generated structured reports were then evaluated by a radiologist for instances of missing or misinterpreted information and any erroneous additional details added by GPT-4. Following this assessment, the template was refined for improved accuracy. This revised template was then used for conversion of 100 other HNCa CT reports into structured format using GPT-4. These reports were then reevaluated in the same manner. Results Initially, GPT-4 successfully converted all 50 free-text reports into structured reports. However, there were 10 places with missing information: tracheostomy tube (n = 3), noninclusion of involvement of sternocleidomastoid muscle (n = 2), extranodal tumor extension (n = 3), and contiguous involvement of the neck structures by nodal mass rather than the primary (n = 2). Few instances of nonsuspicious lung nodules were misinterpreted as metastases (n = 2). GPT-4 did not indicate any erroneous additional findings. Using the revised reporting template, GPT-4 converted all the 100 CT reports into a structured format with no repeated or additional mistakes. Conclusion LLMs can be used for structuring free-text radiology reports using plain language prompts and a simple yet comprehensive reporting template. Summary Statement Large language models can successfully and accurately convert conventional radiology reports for oncology scans into a structured format using a comprehensive predefined template and thus can enhance the utility and integration of these reports in routine clinical practice. Key Points

"Standard" Fractionation for Breast Cancer is No Longer Standard.

This month’s Oncology Scan focuses on 3 recent randomized trials that add to the growing literature on altered fractionation for women with early-stage breast cancer. The Danish and the Chinese experiences study moderate hypofractionation, in which whole breast radiation in the experimental arm is completed within 3 weeks. In each of these studies, hypofractionation proves to be no less effective than the more protracted traditional 5 weeks. Even more convenient is the British 5-fraction regimen, with treatment completed in 1 week. This month’s Oncology Scan summarizes these new trials and offers our perspective on their integration into modern practice.

Decreasing the number of authorization denials in an academic medical oncology practice.

212 Background: Prior authorizations in medical oncology generate additional work and subsequent stress to providers, contributing to physician burnout. Denials of payments can also impact patient care and lead to loss of revenue for the institution. In 2018, first pass denial rates averaged 8.41% per month at our institution. Imaging/scans denials created the majority of this additional work for providers. We aimed to decrease the monthly first pass denial rate average of oncology scans by 25% by May 31, 2020. Methods: Following the creation of a process map of the current prior authorization process and a cause and effect diagram, we identified many factors that could not be controlled (i.e. Insurance company policies). We subsequently created a priority/pay-off-matrix using factors that we could control. Introducing standardized order template for oncology scans was identified as a high impact and feasible countermeasure. Plan-do-study-act cycles (PDSA) plan was developed using this countermeasure to achieve our aim. PDSA 1 included creating template for order entry, educating the nurse partners and advance practice providers (APPs). PDSA 2 included educating physicians and measuring the compliance rate of the template. PDSA 3 addressed the barriers for compliance and education on resources was provided. PDSA 4 included education of the prior authorization staff and reinforcement of template use. A sustainability plan created consisting of a designated RN liaison for review of peer-to-peer requests for oncology scans. Results: PDSA 1: 100% of the nurse partners and APPs were educated. PDSA2: 80% of physicians were educated and 32.8% compliance rate of template use. PDSA 3: 39.2% compliance. PDSA 4: 95% of the radiology prior authorization staff was educated. While our compliance rate for the standardized order template use did increase, we identified many other opportunities to improve the process. Unfortunately, due to COVID-19 pandemic outbreak, resources have been temporarily allocated to relief efforts and the sustainability plan continues to be a work in progress. Conclusions: Peer to peer requests for imaging/scans following authorization denials consume time and effort of providers contributing to burnout and potentially impacting patient care. While many factors cannot be controlled, standardizing ordering process and educating the involved personnel may decrease the number for peer to peer requests. ASCO’s quality training program process helped our institution identify a provider controlled barrier and helped standardize this approach.

Treatment response assessment in [18F]FDG-PET/CT oncology scans: Impact of count statistics variation and reconstruction protocol

PurposeTo investigate influences of reconstruction algorithms and count statistics variation on quantification and treatment response assessment in cancer patients, by using a large field of view-FOV scanner. Methods54 cancer patients underwent PET/CT scan: 1) at baseline: 1.5 min/FOV, reconstructed by ordered-subset expectation maximization + point-spread-function-OSEM-PSF and bayesian penalised-likelihood-BPL algorithm 2) at restaging: 2 min/FOV, reconstructed also at 1.5 and 1 min/FOV, using OSEM-PSF and BPL. SUL (lean-body mass SUV) peak and max were measured for each target-lesion (n = 59). Differences in quantification obtained from datasets with different reconstruction algorithms and different time/FOV were evaluated. For any pair of PET datasets, metabolic response was assessed by using SULpeak, with a threshold of 30% in variation considered as significant. ResultsBoth at baseline and restaging, SULpeak and max values were higher in BPL reconstructions than in OSEM-PSF (p < 0.0001). SULpeak at different time/FOV reconstructions showed no statistically significant differences both with OSEM-PSF and BPL; SULmax depended on acquisition time (p < 0.05). In 56/59 lesions (95%) therapy response was concordant regardless count statistics variation and reconstruction algorithm; 2/59 (3%) showed different responses according to count statistics, both for OSEM-PSF and BPL; in 1/59 lesion (2%) response was different depending on reconstruction algorithm used. ConclusionsBPL provided higher SULpeak and max than OSEM-PSF. With a large FOV/high sensitivity scanner, variation of time/FOV in restaging PET scans gave stable and reproducible results in terms of SULpeak, both for OSEM-PSF and BPL. Thus, metabolic response defined by SULpeak variation proved to be quite independent from count statistics.

Ergebnisse der Österreichischen CT-Dosisstudie 2010: Effektive Dosen der häufigsten CT-Untersuchungen und Unterschiede zwischen Anwendern

PurposeTo determine typical doses from common CT examinations of standard sized adult patients and their variability between CT operators for common CT indications. Materials and MethodsIn a nationwide Austrian CT dose survey doses from approx. 10,000 common CT examinations of adults during 2009 and 2010 were collected and „typical“ radiation doses to the “average patient”, which turned out to have 75.6kg body mass, calculated. Conversion coefficients from DLP to effective dose were determined and effective doses calculated according to ICRP 103. Variations of typically applied doses to the “average patient” were expressed as ratios between 90th and 10th percentile (inter-percentile width, IPW90/10), 1st and 3rd quartile (IPW75/25), and Maximum/Minimum. ResultsMedian effective doses to the average patients for standard head and neck scans were 1.8 mSv (cervical spine), 1.9 mSv (brain: trauma/bleeding, stroke) to 2.2 mSv (brain: masses) with typical variation between facilities of a factor 2.5 (IPW90/10) and 1.7 (IPW75/25). In the thorax region doses were 6.4 to 6.8 mSv (pulmonary embolism, pneumonia and inflammation, oncologic scans), the variation between facilities was by a factor of 2.1 (IPW90/10) and 1.5 (IPW75/25), respectively. In the abdominal region median effective doses from 6.5 mSv (kidney stone search) to 22 mSv (liver lesions) were found (acute abdomen, staging/metastases, lumbar spine: 9-12 mSv; oncologic abdomen plus chest 16 mSv; renal tumor 20 mSv). Variation factors between facilities were on average for abdominal scans 2.7 (IPW90/10) and 1.8 (IPW75/25). ConclusionVariations between CT operators are generally moderate for most operators, but in some indications the ratio between the minimum and the maximum of average dose to the typical standard patients exceeds a factor of 4 or even 5. Therefore, comparing average doses to Diagnostic Reference Levels (DRLs) and optimizing protocols need to be encouraged.

Causes and imaging features of false positives and false negatives on 18F-PET/CT in oncologic imaging

Background18F-FDG is a glucose analogue that is taken up by a wide range of malignancies. 18F-FDG PET-CT is now firmly established as an accurate method for the staging and restaging of various cancers. However, 18F-FDG also accumulates in normal tissue and other non-malignant conditions, and some malignancies do not take up F18-FDG or have a low affinity for the tracer, leading to false-positive and false-negative interpretations.MethodsPET-CT allows for the correlation of two separate imaging modalities, combining both morphological and metabolic information. We should use the CT to help interpret the PET findings. In this article we will highlight specific false-negative and false-positive findings that one should be aware of when interpreting oncology scans.ResultsWe aim to highlight post-treatment conditions that are encountered routinely on restaging scans that can lead to false-positive interpretations. We will emphasise the importance of using the CT component to help recognise these entities to allow improved diagnostic accuracy.ConclusionIn light of the increased use of PET-CT, it is important that nuclear medicine physicians and radiologists be aware of these conditions and correlate the PET and CT components to avoid misdiagnosis, over staging of disease and unnecessary biopsies.

TU‐G‐211‐05: Assessment of Partial Volume Effect after Applying Four Reconstruction Algorithms for SPECT Oncology Scans: Simulation Study

Purpose: We investigated the severity of partial volume effects (PVE) in SPECT oncology studies when images are reconstructed by four different algorithms. Methods: We simulated an oncology SPECT scan with three spherical objects (diameter of 4.4cm, volume of 61mL) representing tumors. These objects were filled with different activities and positioned inside a large cylinder imitating the human body. Our tumor‐like objects were not only surrounded by low‐activity tissues, but also placed next to some organ (for example, liver) with variable activity. We simulated a typical oncology SPECT scan with 60 camera stops over a 360 degree rotation. In total, four methods with increasing complexity were employed to reconstruct images. The first algorithm, M1, included corrections for attenuation and resolution loss and imitated techniques conventionally used in clinics. Method M2 additionally incorporated model‐based scatter correction. Methods M3 and M4 employed the known (from structural imaging modalities) boundaries of tumor(s) and corrected activity inside tumor(s) for spill‐out only (simplified technique M3) or iteratively updated both tumor(s) and background accounting for both spill‐in and spill‐out (advanced technique M4). Results: Incorporation of model‐based scatter correction (method M2) improved the quantification for all tumor‐like objects. Method M1 underestimated tumor activities by 20–25%, whereas method M2 underestimated these activities by only 12–15%. Both PVE corrections M3 and M4 led to substantially better activity estimations with errors of 0–5%. In addition, method M4 provided the best activity distribution in regions encompassing tumors. The relative error of the voxelized activity distribution in that surrounding region was 20–27% for method M1, 19–25% for method M2, 15–18% for method M3, and 12–13% for method M4. Conclusions: The SPECT‐generated activity distribution can be considerably improved by using template‐based corrections for PVE. Correspondingly, the quality of procedures such as tumor staging, estimating response to treatment, and patient‐specific dosimetry can be enhanced.

Quantitative accuracy of the closed-form least-squares solution for targeted SPECT

The aim of this study is to investigate the quantitative accuracy of the closed-form least-squares solution (LSS) for single photon emission computed tomography (SPECT). The main limitation for employing this method in actual clinical reconstructions is the computational cost related to operations with a large-sized system matrix. However, in some clinical situations, the size of the system matrix can be decreased using targeted reconstruction. For example, some oncology SPECT studies are characterized by intense tracer uptakes that are localized in relatively small areas, while the remaining parts of the patient body have only a low activity background. Conventional procedures reconstruct the activity distribution in the whole object, which leads to relatively poor image accuracy/resolution for tumors while computer resources are wasted, trying to rebuild diagnostically useless background. In this study, we apply a concept of targeted reconstruction to SPECT phantom experiments imitating such oncology scans. Our approach includes two major components: (i) disconnection of the entire imaging system of equations and extraction of only those parts that correspond to the targets, i.e., regions of interest (ROI) encompassing active containers/tumors and (ii) generation of the closed-form LSS for each target ROI. We compared these ROI-based LSS with those reconstructed by the conventional MLEM approach. The analysis of the five processed cases from two phantom experiments demonstrated that the LSS approach outperformed MLEM in terms of the noise level inside ROI. On the other hand, MLEM better recovered total activity if the number of iterations was large enough. For the experiment without background activity, the ROI-based LSS led to noticeably better spatial activity distribution inside ROI. However, the distributions pertaining to both approaches were practically identical for the experiment with the concentration ratio 7:1 between the containers and the background.

Iterative reconstruction methods for high-throughput PET tomographs

A fast iterative method is described for processing clinical PET scans acquired in three dimensions, that is, with no inter-plane septa, using standard computers to replace dedicated processors used until the late 1990s. The method is based on sinogram resampling, Fourier rebinning, Monte Carlo scatter simulation and iterative reconstruction using the attenuation-weighted OSEM method and a projector based on a Gaussian pixel model. Resampling of measured sinogram values occurs before Fourier rebinning, to minimize parallax and geometric distortions due to the circular geometry, and also to reduce the size of the sinogram. We analyse the geometrical and statistical effects of resampling, showing that the lines of response are positioned correctly and that resampling is equivalent to about 4 mm of post-reconstruction filtering. We also present phantom and patient results. In this approach, multi-bed clinical oncology scans can be ready for diagnosis within minutes.

Familial caudal regression anomalad and maternal diabetes.

A family is reported which a diabetic woman gave birth to two children with the caudal regression anomalad (CRA). There were no obvious genetic factors. This is the first reported familial case of CRA with maternal diabetes.