Preliminary assessment of automated radiology report generation with generative pre-trained transformers: comparing results to radiologist-generated reports

PURPOSE

In this preliminary study, we aimed to evaluate the potential of the generative pre-trained transformer (GPT) series for generating radiology reports from concise imaging findings and compare its performance with radiologist-generated reports.

METHODS

This retrospective study involved 28 patients who underwent computed tomography (CT) scans and had a diagnosed disease with typical imaging findings. Radiology reports were generated using GPT-2, GPT-3.5, and GPT-4 based on the patient's age, gender, disease site, and imaging findings. We calculated the top-1, top-5 accuracy, and mean average precision (MAP) of differential diagnoses for GPT-2, GPT-3.5, GPT-4, and radiologists. Two board-certified radiologists evaluated the grammar and readability, image findings, impression, differential diagnosis, and overall quality of all reports using a 4-point scale.

RESULTS

Top-1 and Top-5 accuracies for the different diagnoses were highest for radiologists, followed by GPT-4, GPT-3.5, and GPT-2, in that order (Top-1: 1.00, 0.54, 0.54, and 0.21, respectively; Top-5: 1.00, 0.96, 0.89, and 0.54, respectively). There were no significant differences in qualitative scores about grammar and readability, image findings, and overall quality between radiologists and GPT-3.5 or GPT-4 (p > 0.05). However, qualitative scores of the GPT series in impression and differential diagnosis scores were significantly lower than those of radiologists (p < 0.05).

CONCLUSIONS

Our preliminary study suggests that GPT-3.5 and GPT-4 have the possibility to generate radiology reports with high readability and reasonable image findings from very short keywords; however, concerns persist regarding the accuracy of impressions and differential diagnoses, thereby requiring verification by radiologists.

The Feasibility of Using a Deep Learning-Based Model to Determine Cardiac Computed Tomographic Contrast Dose

PURPOSE

This study aimed to predict contrast effects in cardiac computed tomography (CT) from CT localizer radiographs using a deep learning (DL) model and to compare the prediction performance of the DL model with that of conventional models based on patients' physical size.

METHODS

This retrospective study included 473 (256 men and 217 women) cardiac CT scans between May 2014 and August 2017. We developed and evaluated DL models that predict milligrams of iodine per enhancement of the aorta from CT localizer radiographs. To assess the model performance, we calculated and compared Pearson correlation coefficient ( r ) between the actual iodine dose that was necessary to obtain a contrast effect of 1 HU (iodine dose per contrast effect [IDCE]) and IDCE predicted by DL, body weight, lean body weight, and body surface area of patients.

RESULTS

The model was tested on 52 cases for the male group (mean [SD] age, 63.7 ± 11.4) and 44 cases for the female group (mean [SD] age, 69.8 ± 11.6). Correlation coefficients between the actual and predicted IDCE were 0.607 for the male group and 0.412 for the female group, which were higher than the correlation coefficients between the actual IDCE and body weight (0.539 for male, 0.290 for female), lean body weight (0.563 for male, 0.352 for female), and body surface area (0.587 for male, 0.349 for female).

CONCLUSIONS

The performance for predicting contrast effects by analyzing CT localizer radiographs with the DL model was at least comparable with conventional methods using the patient's body size, notwithstanding that no additional measurements other than CT localizer radiographs were required.

Deep learning-based reconstruction and 3D hybrid profile order technique for MRCP at 3T: evaluation of image quality and acquisition time

OBJECTIVES

To evaluate the image quality of the 3D hybrid profile order technique and deep-learning-based reconstruction (DLR) for 3D magnetic resonance cholangiopancreatography (MRCP) within a single breath-hold (BH) at 3 T magnetic resonance imaging (MRI).

METHODS

This retrospective study included 32 patients with biliary and pancreatic disorders. BH images were reconstructed with and without DLR. The signal-to-noise ratio (SNR), contrast, contrast-to-noise ratio (CNR) between the common bile duct (CBD) and periductal tissues, and full width at half maximum (FWHM) of CBD on 3D-MRCP were evaluated quantitatively. Two radiologists scored image noise, contrast, artifacts, blur, and overall image quality of the three image types using a 4-point scale. Quantitative and qualitative scores were compared using the Friedman test and post hoc Nemenyi test.

RESULTS

The SNR and CNR were not significantly different when under respiratory gating- and BH-MRCP without DLR. However, they were significantly higher under BH with DLR than under respiratory gating (SNR, p = 0.013; CNR, p = 0.027). The contrast and FWHM of MRCP under BH with and without DLR were lower than those under respiratory gating (contrast, p < 0.001; FWHM, p = 0.015). Qualitative scores for noise, blur, and overall image quality were higher under BH with DLR than those under respiratory gating (blur, p = 0.003; overall, p = 0.008).

CONCLUSIONS

The combination of the 3D hybrid profile order technique and DLR is useful for MRCP within a single BH and does not lead to the deterioration of image quality and space resolution at 3 T MRI.

CLINICAL RELEVANCE STATEMENT

Considering its advantages, this sequence might become the standard protocol for MRCP in clinical practice, at least at 3.0 T.

KEY POINTS

• The 3D hybrid profile order can achieve MRCP within a single breath-hold without a decrease in spatial resolution. • The DLR significantly improved the CNR and SNR of BH-MRCP. • The 3D hybrid profile order technique with DLR reduces the deterioration of image quality in MRCP within a single breath-hold.

Ramucirumab plus FOLFIRI as second-line treatment for patients with RAS wild-type metastatic colorectal cancer previously treated with anti-EGFR antibody: JACCRO CC-16

BACKGROUND

Chemotherapy in combination with anti-epidermal growth factor receptor (EGFR) antibody is considered a first-line treatment regimen for RAS wild-type and left-sided metastatic colorectal cancer (mCRC), whereas second-line treatment regimens have not yet been established. Few studies have prospectively evaluated second-line treatment with anti-vascular endothelial growth factor antibody after first-line anti-EGFR antibody therapy for RAS wild-type mCRC.

PATIENTS AND METHODS

This non-randomized phase II trial investigated the clinical outcomes of second-line ramucirumab (RAM) plus fluorouracil, levofolinate, and irinotecan (FOLFIRI) after first-line anti-EGFR antibody in combination with doublet or triplet regimen in patients with RAS wild-type mCRC. The primary endpoint was the 6-month progression-free survival (PFS) rate. The secondary endpoints were PFS, overall survival (OS), objective response rate (ORR), rate of early tumor shrinkage (ETS), and safety. We hypothesized a threshold 6-month PFS rate of 30% and an expected 6-month PFS rate of 45%. Treatment was considered effective if the lower limit of the 90% confidence interval (CI) of the 6-month PFS rate was >0.30.

RESULTS

Ninety-two patients were enrolled in the study. The primary tumor was located on the left side in 86 (95.6%) patients. Twenty (22.0%) patients had received triplet plus cetuximab as previous therapy. Six-month PFS rate was 58.2% (90% CI 49.3% to 66.2%) with a median PFS of 7.0 months (95% CI 5.7-7.6 months). Median OS was 23.6 months (95% CI 16.5-26.3 months). The ORR and ETS rate were 10.7% and 16.9%, respectively, in 83 patients with measurable lesions. The 6-month PFS rate was comparable between patients previously treated with doublet and triplet regimens; however, median PFS was longer for the doublet regimen (7.4 versus 6.4 months, P = 0.036).

CONCLUSIONS

Our study demonstrated prospectively that RAM plus FOLFIRI is an effective second-line treatment after anti-EGFR antibody-containing first-line therapy in RAS wild-type and left-sided mCRC. Furthermore, the results were similar for patients who were previously treated with triplet regimen.

Combination Use of Compressed Sensing and Deep Learning for Shoulder Magnetic Resonance Imaging With Various Sequences

OBJECTIVE

For compressed sensing (CS) to become widely used in routine magnetic resonance imaging (MRI), it is essential to improve image quality. This study aimed to evaluate the usefulness of combining CS and deep learning-based reconstruction (DLR) for various sequences in shoulder MRI.

METHODS

This retrospective study included 37 consecutive patients who underwent undersampled shoulder MRIs, including T1-weighted (T1WI), T2-weighted (T2WI), and fat-saturation T2-weighted (FS-T2WI) images. Images were reconstructed using the conventional wavelet-based denoising method (wavelet method) and a combination of wavelet and DLR-based denoising methods (hybrid-DLR method) for each sequence. The signal-to-noise ratio and contrast-to-noise ratio of the bone, muscle, and fat and the full width at half maximum of the shoulder joint were compared between the 2 image types. In addition, 2 board-certified radiologists scored the image noise, contrast, sharpness, artifacts, and overall image quality of the 2 image types on a 4-point scale.

RESULTS

The signal-to-noise ratios and contrast-to-noise ratios of the bone, muscle, and fat in T1WI, T2WI, and FS-T2WI obtained from the hybrid-DLR method were significantly higher than those of the conventional wavelet method (P < 0.001). However, there were no significant differences in the full width at half maximum of the shoulder joint in any of the sequences (P > 0.05). Furthermore, in all sequences, the mean scores of the image noise, sharpness, artifacts, and overall image quality were significantly higher in the hybrid-DLR method than in the wavelet method (P < 0.001), but there were no significant differences in contrast among the sequences (P > 0.05).

CONCLUSIONS

The DLR denoising method can improve the image quality of CS in T1-weighted images, T2-weighted images, and fat-saturation T2-weighted images of the shoulder compared with the wavelet denoising method alone.

Search for the Majorana Nature of Neutrinos in the Inverted Mass Ordering Region with KamLAND-Zen

The KamLAND-Zen experiment has provided stringent constraints on the neutrinoless double-beta (0νββ) decay half-life in ^{136}Xe using a xenon-loaded liquid scintillator. We report an improved search using an upgraded detector with almost double the amount of xenon and an ultralow radioactivity container, corresponding to an exposure of 970 kg yr of ^{136}Xe. These new data provide valuable insight into backgrounds, especially from cosmic muon spallation of xenon, and have required the use of novel background rejection techniques. We obtain a lower limit for the 0νββ decay half-life of T_{1/2}^{0ν}>2.3×10^{26}  yr at 90% C.L., corresponding to upper limits on the effective Majorana neutrino mass of 36-156 meV using commonly adopted nuclear matrix element calculations.

Update on the Use of Artificial Intelligence in Hepatobiliary MR Imaging

The application of machine learning (ML) and deep learning (DL) in radiology has expanded exponentially. In recent years, an extremely large number of studies have reported about the hepatobiliary domain. Its applications range from differential diagnosis to the diagnosis of tumor invasion and prediction of treatment response and prognosis. Moreover, it has been utilized to improve the image quality of DL reconstruction. However, most clinicians are not familiar with ML and DL, and previous studies about these concepts are relatively challenging to understand. In this review article, we aimed to explain the concepts behind ML and DL and to summarize recent achievements in their use in the hepatobiliary region.

POS0687 INHIBITION OF BONE EROSION, DETERMINED BY HIGH-RESOLUTION PERIPHERAL QUANTITATIVE COMPUTED TOMOGRAPHY (HR-pQCT), IN RHEUMATOID ARTHRITIS PATIENTS RECEIVING A CONVENTIONAL SYNTHETIC DMARD (csDMARD) PLUS DENOSUMAB VS csDMARD THERAPY ALONE: RESULTS OF AN OPEN-LABEL, RANDOMIZED, PARALLEL-GROUP STUDY

Background

Denosumab, a human IgG2 monoclonal antibody with high affinity for RANKL, is approved for treatment of bone erosion (ER) in patients with RA, based on the results of clinical trials. However, its effectiveness in combination with conventional therapy in RA patients has not been fully investigated in clinical practice.

Objectives

This exploratory study aimed to compare, in patients receiving conventional synthetic disease-modifying anti-rheumatic drugs (csDMARD) for treatment of RA, the effectiveness of combined use of csDMARD and denosumab vs csDMARD alone, in terms of inhibition of ER determined by HR-pQCT.

Methods

Detail protocol of this open-label, randomized, parallel-group study has been published previously.1 RA patients with moderate or low disease activity and progressive ER were eligible, and were randomly assigned to receive denosumab in addition to the csDMARD (denosumab) group or to continued use of the csDMARD alone (csDMARD) group. Denosumab was administered every 6 months during the 12-month study period. The primary endpoint was change in ER-depth at the second and third metacarpal bones, determined by HR-pQCT at month 6. For the primary endpoint, a linear mixed effect model analysis was performed using treatment group, sex, anti-cyclic citrullinated peptide (CCP) antibody (positive vs negative), and baseline disease activity (DAS28-ESR) as fixed effects, patients as random effects, and baseline values as covariates. For extension data, measurement time-point and the interaction between treatment group and measurement time-point was further added as fixed effects. The adverse events (AEs) were recorded.

Results

A total of 46 patients were randomized to denosumab and csDMARD groups (both N=23), and baseline characteristics were similar between both groups. Although the difference was not significant, the estimate mean (95%CI) change of ER-depth at month 6 from baseline as the primary endpoint was −0.57 (−1.52, 0.39) in the denosumab group vs −0.22 (−0.97, 0.53) in the csDMARD group, respectively. At months 6 and 12, ER-depth, -width, and -volume (extension data) were numerically lower in the denosumab group than in the csDMARD group (Table 1). Compared with patients in the csDMARD group, those in the denosumab group had significantly higher volumetric bone mineral density (vBMD) at month 12. AEs were reported in 12 (52.2%) and 13 (56.5%) of patients in the denosumab and csDMARD groups, respectively. The most common AEs of denosumab groups was hypocalcemia (reported in 4), and all the events were causally related with denosumab. Serious AEs were reported in 1 (4.3%) and 2 (8.7%) of patients in the denosumab and csDMARD groups, and which were not related to treatment drug.

Table 1.

ER and microstructure in denosumab group vs csDMARD group

MonthDenosumab group (n=21)csDMARDs group (n=22)Difference (Denosumab- csDMARDs)n1Estimate Means [95%CI]n1Estimate Means [95%CI]Estimate Means [95%CI]ER-depth617−0.46 [−1.31, 0.39]25−0.20 [−0.89, 0.49]−0.27 [−0.86, 0.32]1217−0.56 [−1.41, 0.29]22−0.20 [−0.90, 0.49]−0.35 [−0.95, 0.24]ER-width617−0.26 [−1.10, 0.57]25−0.06 [−0.73, 0.61]−0.20 [−0.81, 0.40]1217−0.27 [−1.10, 0.56]22−0.03 [−0.70, 0.64]−0.24 [−0.85, 0.38]ER-volume616−6.21 [−23.89, 11.46]24−1.71 [−16.07, 12.66]−4.51 [−16.67, 7.65]1217−6.25 [−23.94, 11.44]21−3.18 [−17.56, 11.20]−3.07 [−15.32, 9.17]vBMD6426.41 [2.78, 10.03]442.46 [−1.33, 6.24]3.95 [−0.05, 7.94]12389.20 [5.46, 12.95]423.66 [−0.15, 7.46]5.55 [1.46, 9.63]*

These were extension data.

This is the efficacy analysis set which excluded 3 patients from the full analysis set.

*; p<0.01

n, number of patients; n1, number of measurement points

Conclusion

Our results suggest that adding denosumab to csDMARD therapy may help prevent ER, promote ER repair, and improve bone microstructure.

References

[1]Iwamoto N, et al., Trials. 2019;20(1):1–8.

Disclosure of Interests

Naoki Iwamoto Speakers bureau: Daiichi Sankyo Co., Ltd., Grant/research support from: Daiichi Sankyo Co., Ltd., Ko Chiba Speakers bureau: Daiichi Sankyo Co., Ltd., Shuntaro Sato: None declared, Kazuteru Shiraishi: None declared, Konosuke Watanabe: None declared, Nozomi Ohki: None declared, Akitomo Okada: None declared, Tomohiro Koga: None declared, Makiko Kobayashi Employee of: Daiichi Sankyo Co., Ltd. (retired at submission), Kengo Saito Employee of: Daiichi Sankyo Co., Ltd., Naoki Okubo Employee of: Daiichi Sankyo Co., Ltd., Atsushi Kawakami Speakers bureau: Daiichi Sankyo Co., Ltd., Grant/research support from: Daiichi Sankyo Co., Ltd.

The utility of patent ductus arteriosus closure with hemostatic clip in dogs

This study investigated the utility of patent ductus arteriosus (PDA) closure with hemostatic clip by comparing with traditional PDA closure. Medical records of 51 dogs with surgical closure of PDA were reviewed and retrospective study was conducted. 29 dogs were treated by procedure with hemostatic clip (Group HC), and 22 dogs were treated by surgical ligation (Group SL). Data pertaining to breed, sex, age and body weight at the time of surgery, echocardiographic minimal ductal diameter, duration of surgery, hemostatic clip size, echocardiographic findings, hemor-rhage, residual ductal flow and recanalization were collected from records. The results showed that procedure with hemostatic clip had been selected in lighter dogs than traditional PDA closure. Duration of surgery performed only hemostatic clip technique was significantly shorter than that in group SL. Preoperative LVIDd, E-wave and FS were significantly lower than postoperative ones. As regard all parameters, the differences between pre- and postoperative periods were not significantly different between group HC and group SL. Hemorrhage, residual ductal flow, and recanalization were not significantly different in both groups. The present study showed that procedure with hemostatic clip is beneficial in that it is available in smaller dogs and can make shorter operation duration than traditional PDA closure. Moreover, the procedure is effective for the resolution of volume overload of the left atrium and ventricle in short-term outcome. Complications including hemorrhage, residual ductal flow and recanaliza-tion were not significantly different with both techniques.

Search for η^{'} Bound Nuclei in the ^{12}C(γ,p) Reaction with Simultaneous Detection of Decay Products

We measured missing mass spectrum of the ^{12}C(γ,p) reaction for the first time in coincidence with potential decay products from η^{'} bound nuclei. We tagged an (η+p) pair associated with the η^{'}N→ηN process in a nucleus. After applying kinematical selections to reduce backgrounds, no signal events were observed in the bound-state region. An upper limit of the signal cross section in the opening angle cosθ_{lab}^{ηp}<-0.9 was obtained to be 2.2  nb/sr at the 90% confidence level. It is compared with theoretical cross sections, whose normalization ambiguity is suppressed by measuring a quasifree η^{'} production rate. Our results indicate a small branching fraction of the η^{'}N→ηN process and/or a shallow η^{'}-nucleus potential.

Fluorescence Database of Aerosol-Candidate-Substances for Fluorescence Lidar Application

A database containing spectrum and cross-section of the fluorescence of substances has been made. In test of forest environment monitoring by our Laser-Induced Fluorescence Spectrum (LIFS) lidar, the database showed that the origin substance of the aerosol observed by the lidar was cedar pollen, and the concentration was calculated using the cross-section. In the urban atmosphere monitoring, three substances stored in the database were proposed to be the origins of the aerosol. Based on these experiments, we discuss the usefulness of the fluorescence database in lidar observations.

Applicability of MIL-101(Fe) as a cathode of lithium ion batteries

MIL-101(Fe) was investigated as a cathode material of lithium ion batteries. A battery test reveals that MIL-101(Fe) shows a charge and discharge capacitance of 110 mA h g⁻¹.