Comparing Diagnostic Accuracy of Radiologists versus GPT-4V and Gemini Pro Vision Using Image Inputs from Diagnosis Please Cases.

Background The diagnostic abilities of multimodal large language models (LLMs) using direct image inputs and the impact of the temperature parameter of LLMs remain unexplored. Purpose To investigate the ability of GPT-4V and Gemini Pro Vision in generating differential diagnoses at different temperatures compared with radiologists using Radiology Diagnosis Please cases. Materials and Methods This retrospective study included Diagnosis Please cases published from January 2008 to October 2023. Input images included original images and captures of the textual patient history and figure legends (without imaging findings) from PDF files of each case. The LLMs were tasked with providing three differential diagnoses, repeated five times at temperatures 0, 0.5, and 1. Eight subspecialty-trained radiologists solved cases. An experienced radiologist compared generated and final diagnoses, considering the result correct if the generated diagnoses included the final diagnosis after five repetitions. Accuracy was assessed across models, temperatures, and radiology subspecialties, with statistical significance set at P < .007 after Bonferroni correction for multiple comparisons across the LLMs at the three temperatures and with radiologists. Results A total of 190 cases were included in neuroradiology (n = 53), multisystem (n = 27), gastrointestinal (n = 25), genitourinary (n = 23), musculoskeletal (n = 17), chest (n = 16), cardiovascular (n = 12), pediatric (n = 12), and breast (n = 5) subspecialties. Overall accuracy improved with increasing temperature settings (0, 0.5, 1) for both GPT-4V (41% [78 of 190 cases], 45% [86 of 190 cases], 49% [93 of 190 cases], respectively) and Gemini Pro Vision (29% [55 of 190 cases], 36% [69 of 190 cases], 39% [74 of 190 cases], respectively), although there was no evidence of a statistically significant difference after Bonferroni adjustment (GPT-4V, P = .12; Gemini Pro Vision, P = .04). The overall accuracy of radiologists (61% [115 of 190 cases]) was higher than that of Gemini Pro Vision at temperature 1 (T1) (P < .001), while no statistically significant difference was observed between radiologists and GPT-4V at T1 after Bonferroni adjustment (P = .02). Radiologists (range, 45%-88%) outperformed the LLMs at T1 (range, 24%-75%) in most subspecialties. Conclusion Using direct radiologic image inputs, GPT-4V and Gemini Pro Vision showed improved diagnostic accuracy with increasing temperature settings. Although GPT-4V slightly underperformed compared with radiologists, it nonetheless demonstrated promising potential as a supportive tool in diagnostic decision-making. © RSNA, 2024 See also the editorial by Nishino and Ballard in this issue.

Diagnosis of Unruptured Intracranial Aneurysms Using Proton-Density Magnetic Resonance Angiography: A Comparison With High-Resolution Time-of-Flight Magnetic Resonance Angiography.

OBJECTIVE: Differentiating intracranial aneurysms from normal variants using CT angiography (CTA) or MR angiography (MRA) poses significant challenges. This study aimed to evaluate the efficacy of proton-density MRA (PD-MRA) compared to high-resolution time-of-flight MRA (HR-MRA) in diagnosing aneurysms among patients with indeterminate findings on conventional CTA or MRA. MATERIALS AND METHODS: In this retrospective analysis, we included patients who underwent both PD-MRA and HR-MRA from August 2020 to July 2022 to assess lesions deemed indeterminate on prior conventional CTA or MRA examinations. Three experienced neuroradiologists independently reviewed the lesions using HR-MRA and PD-MRA with reconstructed voxel sizes of 0.253 mm3 or 0.23 mm3, respectively. A neurointerventionist established the gold standard with digital subtraction angiography. We compared the performance of HR-MRA, PD-MRA (0.253-mm3 voxel), and PD-MRA (0.23-mm3 voxel) in diagnosing aneurysms, both per lesion and per patient. The Fleiss kappa statistic was used to calculate inter-reader agreement. RESULTS: The study involved 109 patients (average age 57.4 ± 11.0 years; male:female ratio, 11:98) with 141 indeterminate lesions. Of these, 78 lesions (55.3%) in 69 patients were confirmed as aneurysms by the reference standard. PD-MRA (0.253-mm3 voxel) exhibited significantly higher per-lesion diagnostic performance compared to HR-MRA across all three readers: sensitivity ranged from 87.2%-91.0% versus 66.7%-70.5%; specificity from 93.7%-96.8% versus 58.7%-68.3%; and accuracy from 90.8%-92.9% versus 63.8%-69.5% (P ≤ 0.003). Furthermore, PD-MRA (0.253-mm3 voxel) demonstrated significantly superior per-patient specificity and accuracy compared to HR-MRA across all evaluators (P ≤ 0.013). The diagnostic accuracy of PD-MRA (0.23-mm3 voxel) surpassed that of HR-MRA and was comparable to PD-MRA (0.253-mm3 voxel). The kappa values for inter-reader agreements were significantly higher in PD-MRA (0.820-0.938) than in HR-MRA (0.447-0.510). CONCLUSION: PD-MRA outperformed HR-MRA in diagnostic accuracy and demonstrated almost perfect inter-reader consistency in identifying intracranial aneurysms among patients with lesions initially indeterminate on CTA or MRA.

Ethanol Ablation of Ranulas and Risk Factor Analysis for Recurrence.

Importance: Ethanol ablation (EA) was shown to be safe and effective for treating ranula, but few studies have assessed long-term outcomes and recurrence of ranula after EA. Objective: To evaluate the long-term outcomes and the risk factors for recurrence and receipt of subsequent surgery in patients who underwent treatment with EA for ranula. Design, Setting, and Participants: This case-series study was conducted at a single tertiary hospital and assessed patients who were treated with EA between July 2009 and March 2021. Among 70 consecutive patients, those with follow-up loss or who were followed up for less than 24 months were excluded. Exposures: EA for ranula. Main Outcomes and Measures: The primary outcome was recurrence at last follow-up after single or multiple EA sessions. Secondary outcomes included receipt of subsequent surgery and the recurrence-free survival (RFS) rate after initial EA. Factors possibly associated with outcomes included patient age and sex; ranula site, type, diameter, volume, and echogenicity; the presentation-to-EA interval; parapharyngeal space extension; and sublingual gland herniation. Risk factors were identified on logistic regression analyses. Two-year RFS rates were analyzed for the initial cohort using the Kaplan-Meier method and compared by log-rank tests. Results: A total of 57 patients (mean [SD] age, 26.4 [12.1] years; 24 female individuals [42%]) who were followed up for a median of 57 months (range, 24-167 months) were included. The recurrence rate was 33% (n = 19), and 11 (19%) underwent subsequent surgery. Among patients with recurrence, 86% (31 of 36) experienced first recurrence within 12 months after initial EA. A presentation-to-EA interval of 12 months or longer was associated with an increased risk of recurrence (adjusted odds ratio [OR], 3.74; 95% CI, 1.01-13.82). No risk factors were significantly associated with subsequent surgery (highest OR in parapharyngeal space extension: adjusted OR, 4.96; 95% CI, 0.94-26.35). Among the initial cohort of 70 patients, 2-year RFS was lower in a maximum diameter of ranula of 5 cm or greater than less than 5 cm (24% [95% CI, 7%-41%] vs 50% [95% CI, 34%-66%]; difference, 26% [95% CI, -4% to 56%]; log-rank test, P = .02). Conclusions and Relevance: This case-series study found that the recurrence rate of ranula after EA was 33%. A presentation-to-EA interval of 12 months or longer may be a risk factor for recurrence, suggesting that early intervention with EA might minimize recurrence. Most first recurrences occurred within 12 months after EA, with a maximum diameter of ranula of 5 cm or greater being a possible risk factor.

Improving Diagnostic Performance of MRI for Temporal Lobe Epilepsy With Deep Learning-Based Image Reconstruction in Patients With Suspected Focal Epilepsy.

OBJECTIVE: To evaluate the diagnostic performance and image quality of 1.5-mm slice thickness MRI with deep learning-based image reconstruction (1.5-mm MRI + DLR) compared to routine 3-mm slice thickness MRI (routine MRI) and 1.5-mm slice thickness MRI without DLR (1.5-mm MRI without DLR) for evaluating temporal lobe epilepsy (TLE). MATERIALS AND METHODS: This retrospective study included 117 MR image sets comprising 1.5-mm MRI + DLR, 1.5-mm MRI without DLR, and routine MRI from 117 consecutive patients (mean age, 41 years; 61 female; 34 patients with TLE and 83 without TLE). Two neuroradiologists evaluated the presence of hippocampal or temporal lobe lesions, volume loss, signal abnormalities, loss of internal structure of the hippocampus, and lesion conspicuity in the temporal lobe. Reference standards for TLE were independently constructed by neurologists using clinical and radiological findings. Subjective image quality, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) were analyzed. Performance in diagnosing TLE, lesion findings, and image quality were compared among the three protocols. RESULTS: The pooled sensitivity of 1.5-mm MRI + DLR (91.2%) for diagnosing TLE was higher than that of routine MRI (72.1%, P < 0.001). In the subgroup analysis, 1.5-mm MRI + DLR showed higher sensitivity for hippocampal lesions than routine MRI (92.7% vs. 75.0%, P = 0.001), with improved depiction of hippocampal T2 high signal intensity change (P = 0.016) and loss of internal structure (P < 0.001). However, the pooled specificity of 1.5-mm MRI + DLR (76.5%) was lower than that of routine MRI (89.2%, P = 0.004). Compared with 1.5-mm MRI without DLR, 1.5-mm MRI + DLR resulted in significantly improved pooled accuracy (91.2% vs. 73.1%, P = 0.010), image quality, SNR, and CNR (all, P < 0.001). CONCLUSION: The use of 1.5-mm MRI + DLR enhanced the performance of MRI in diagnosing TLE, particularly in hippocampal evaluation, because of improved depiction of hippocampal abnormalities and enhanced image quality.

Knot-and-ear sign: a pathognomonic ultrasonographic feature of suture granuloma after thyroid surgery.

PURPOSE: This study investigated the clinical and ultrasonographic (US) findings of suture granulomas and recurrent tumors, and aimed to identify specific characteristics of suture granulomas through an experimental study. METHODS: This retrospective study included 20 pathologically confirmed suture granulomas and 40 recurrent tumors between January 2010 and December 2020. The clinical findings included suture material, surgery, and initial TNM stage. The US findings included shape, size, margin, echogenicity, heterogeneity, vascularity, and internal echogenic foci. The distribution, paired appearance, and "knot-and-ear" appearance of internal echogenic foci were assessed. An experiment using pork meat investigated the US configuration of suture knots. RESULTS: Eighteen patients with 20 suture granulomas (15 women; mean age, 52±13 years) and 37 patients with 40 recurrent tumors (24 women; 54±18 years) were included. Patients with suture granulomas exhibited earlier initial T and N stages than those with recurrent tumors. The knot-and-ear appearance, defined as an echogenic dot accompanied by two adjacent echogenic dots or lines based on experimental findings, was observed in 50% of suture granulomas, but not in recurrent tumors (P<0.001). Central internal echogenic foci (68.8%, P=0.023) and paired appearance (75.0%, P<0.001) were more frequent in suture granulomas. During follow-up, 94.1% of suture granulomas shrunk. CONCLUSION: The knot-and-ear appearance is a potential pathognomonic finding of suture granuloma, and central internal echogenic foci with paired appearance were typical US features. Clinically, suture granulomas showed early T and N stages and size reduction during follow-up. Understanding these features can prevent unnecessary biopsy or diagnostic surgery.

Corrigendum: Development and validation of a deep learning-based automatic segmentation model for assessing intracranial volume: comparison with NeuroQuant, FreeSurfer, and SynthSeg.

[This corrects the article DOI: 10.3389/fneur.2023.1221892.].

Impact of a selective lens dose reduction protocol in 3D rotational angiography on radiation exposure to the eye lens during cerebral angiography: a randomized controlled trial.

BACKGROUND: We aimed to investigate the radiation dose to the eye lens (lens dose) during cerebral angiography and to evaluate the effectiveness of the lens dose reduction protocol for 3-dimensional rotational angiography (3D-RA) in reducing overall lens dose exposure. METHODS: We conducted a randomized, controlled clinical trial at a tertiary hospital with patients undergoing cerebral angiography. The lens dose reduction protocol in 3D-RA involved raising the table to position the patient's eye lens away from the rotation axis. The lens dose was estimated by measuring the entrance surface air kerma using a photoluminescent glass dosimeter. The lens doses of 3D-RA, overall examination, and image quality were analyzed and compared between the two groups. RESULTS: A total of 20 participants (mean age, 58±9.4 years; including 12 men [60%]) were enrolled and randomly assigned to either the conventional group or the dose reduction group. The median lens dose in 3D-RA was significantly lower in the dose reduction group compared with the conventional group (1.1 mGy vs 4.5 mGy, p<0.001). The total dose was significantly lower in the dose reduction group (median of 7.5 mGy vs 10.2 mGy, p=0.003). In the conventional group, 3D-RA accounted for 46% of the total lens dose, while in the dose reduction group, its proportion decreased to 16%. No significant differences were observed in the image quality between the groups. CONCLUSION: The lens dose reduction protocol resulted in a significant reduction in the lens dose of the 3D-RA as well as entire cerebral angiography, while maintaining the image quality.

Development and validation of a deep learning-based automatic segmentation model for assessing intracranial volume: comparison with NeuroQuant, FreeSurfer, and SynthSeg.

Background and purpose: To develop and validate a deep learning-based automatic segmentation model for assessing intracranial volume (ICV) and to compare the accuracy determined by NeuroQuant (NQ), FreeSurfer (FS), and SynthSeg. Materials and methods: This retrospective study included 60 subjects [30 Alzheimer's disease (AD), 21 mild cognitive impairment (MCI), 9 cognitively normal (CN)] from a single tertiary hospital for the training and validation group (50:10). The test group included 40 subjects (20 AD, 10 MCI, 10 CN) from the ADNI dataset. We propose a robust ICV segmentation model based on the foundational 2D UNet architecture trained with four types of input images (both single and multimodality using scaled or unscaled T1-weighted and T2-FLAIR MR images). To compare with our model, NQ, FS, and SynthSeg were also utilized in the test group. We evaluated the model performance by measuring the Dice similarity coefficient (DSC) and average volume difference. Results: The single-modality model trained with scaled T1-weighted images showed excellent performance with a DSC of 0.989 ± 0.002 and an average volume difference of 0.46% ± 0.38%. Our multimodality model trained with both unscaled T1-weighted and T2-FLAIR images showed similar performance with a DSC of 0.988 ± 0.002 and an average volume difference of 0.47% ± 0.35%. The overall average volume difference with our model showed relatively higher accuracy than NQ (2.15% ± 1.72%), FS (3.69% ± 2.93%), and SynthSeg (1.88% ± 1.18%). Furthermore, our model outperformed the three others in each subgroup of patients with AD, MCI, and CN subjects. Conclusion: Our deep learning-based automatic ICV segmentation model showed excellent performance for the automatic evaluation of ICV.

Reduction of Radiation Dose to Eye Lens in Cerebral 3D Rotational Angiography Using Head Off-Centering by Table Height Adjustment: A Prospective Study.

OBJECTIVE: Three-dimensional rotational angiography (3D-RA) is increasingly used for the evaluation of intracranial aneurysms (IAs); however, radiation exposure to the lens is a concern. We investigated the effect of head off-centering by adjusting table height on the lens dose during 3D-RA and its feasibility in patient examination. MATERIALS AND METHODS: The effect of head off-centering during 3D-RA on the lens radiation dose at various table heights was investigated using a RANDO head phantom (Alderson Research Labs). We prospectively enrolled 20 patients (58.0 ± 9.4 years) with IAs who were scheduled to undergo bilateral 3D-RA. In all patients' 3D-RA, the lens dose-reduction protocol involving elevation of the examination table was applied to one internal carotid artery, and the conventional protocol was applied to the other. The lens dose was measured using photoluminescent glass dosimeters (GD-352M, AGC Techno Glass Co., LTD), and radiation dose metrics were compared between the two protocols. Image quality was quantitatively analyzed using source images for image noise, signal-to-noise ratio, and contrast-to-noise ratio. Additionally, three reviewers qualitatively assessed the image quality using a five-point Likert scale. RESULTS: The phantom study showed that the lens dose was reduced by an average of 38% per 1 cm increase in table height. In the patient study, the dose-reduction protocol (elevating the table height by an average of 2.3 cm) led to an 83% reduction in the median dose from 4.65 mGy to 0.79 mGy (P < 0.001). There were no significant differences between dose-reduction and conventional protocols in the kerma area product (7.34 vs. 7.40 Gy·cm², P = 0.892), air kerma (75.7 vs. 75.1 mGy, P = 0.872), and image quality. CONCLUSION: The lens radiation dose was significantly affected by table height adjustment during 3D-RA. Intentional head off-centering by elevation of the table is a simple and effective way to reduce the lens dose in clinical practice.

Three cases of pancreatic pseudocysts associated with dorsal pancreatic agenesis.

Agenesis of the dorsal pancreas (ADP) is an extremely rare congenital anomaly. Human pancreas is formed by ventral and dorsal endodermal buds of the foregut endoderm. The dorsal bud forms the upper part of the head, neck, body, and tail of the pancreas and the ventral bud generates most of the head and uncinate process. ADP is derived from the embryologic failure of the dorsal pancreatic bud to form the pancreatic body and tail. ADP can be related to some diseases and conditions such as pancreatitis, hypoglycemia, and rarely pancreatic tumors. The association between cystic lesions with ADP has previously been reported. Three cases of cystic lesions of the pancreas with ADP were diagnosed clinically based on the imaging features and without any past history of pancreatitis. However, the pathologic diagnosis of resected lesions confirmed pseudocysts without pathologic evidence of tumor. We report 3 cases of pancreatic pseudocysts associated with ADP.

Effects of Smoking on Menopausal Age: Results From the Korea National Health and Nutrition Examination Survey, 2007 to 2012.

OBJECTIVES: Decreased fertility and impaired health owing to early menopause are significant health issues. Smoking is a modifiable health-related behavior that influences menopausal age. We investigated the effects of smoking-associated characteristics on menopausal age in Korean women. METHODS: This study used data from the Korea National Health and Nutrition Examination Survey from 2007 to 2012. Menopausal age in relation to smoking was analyzed as a Kaplan-Meier survival curve for 11 510 women (aged 30 to 65 years). The risk of entering menopause and experiencing early menopause (before age 48) related to smoking were assessed using a Cox proportional hazards model. RESULTS: The menopausal age among smokers was 0.75 years lower than that among non-smokers (p<0.001). The results of the Cox proportional hazards model showed pre-correction and post-correction risk ratios for entering menopause related to smoking of 1.26 (95% confidence interval [CI], 1.09 to 1.46) and 1.27 (95% CI, 1.10 to 1.47), respectively, and pre-correction and post-correction risk ratios for experiencing early menopause related to smoking of 1.36 (95% CI, 1.03 to 1.80) and 1.40 (95% CI, 1.05 to 1.85), respectively. CONCLUSIONS: Smokers reached menopause earlier than non-smokers, and their risk for experiencing early menopause was higher.