Performance of GPT-4 on the American College of Radiology In-training Examination: Evaluating Accuracy, Model Drift, and Fine-tuning.

RATIONALE AND OBJECTIVES: In our study, we evaluate GPT-4's performance on the American College of Radiology (ACR) 2022 Diagnostic Radiology In-Training Examination (DXIT). We perform multiple experiments across time points to assess for model drift, as well as after fine-tuning to assess for differences in accuracy. MATERIALS AND METHODS: Questions were sequentially input into GPT-4 with a standardized prompt. Each answer was recorded and overall accuracy was calculated, as was logic-adjusted accuracy, and accuracy on image-based questions. This experiment was repeated several months later to assess for model drift, then again after the performance of fine-tuning to assess for changes in GPT's performance. RESULTS: GPT-4 achieved 58.5% overall accuracy, lower than the PGY-3 average (61.9%) but higher than the PGY-2 average (52.8%). Adjusted accuracy was 52.8%. GPT-4 showed significantly higher (p = 0.012) confidence for correct answers (87.1%) compared to incorrect (84.0%). Performance on image-based questions was significantly poorer (p < 0.001) at 45.4% compared to text-only questions (80.0%), with adjusted accuracy for image-based questions of 36.4%. When the questions were repeated, GPT-4 chose a different answer 25.5% of the time and there was no change in accuracy. Fine-tuning did not improve accuracy. CONCLUSION: GPT-4 performed between PGY-2 and PGY-3 levels on the 2022 DXIT, significantly poorer on image-based questions, and with large variability in answer choices across time points. Exploratory experiments in fine-tuning did not improve performance. This study underscores the potential and risks of using minimally-prompted general AI models in interpreting radiologic images as a diagnostic tool. Implementers of general AI radiology systems should exercise caution given the possibility of spurious yet confident responses.

Increased Vertebral Gallium-68-DOTATATE Activity on Positron Emission Tomography-Computed Tomography in Merkel Cell Carcinoma: Not Always Metastasis.

Merkel cell carcinoma (MCC) is a rare neuroendocrine dermal malignancy seen in elderly light-skinned individuals, associated with immunosuppression and Merkel cell polyomavirus infection. As a neuroendocrine tumor, the recurrence and metastasis of MCC can be evaluated using positron emission tomography-computed tomography (PET-CT) with the Gallium-68-DOTATATE (Ga-68-DOTATATE) radiotracer, which has demonstrated increased sensitivity to neuroendocrine metastases when compared to F-18 fluorodeoxyglucose (FDG). Here, we present the case of a patient with known metastatic MCC with a new, abnormal focus of increased radiotracer activity in the thoracic spine on Ga-68-DOTATATE PET-CT suspected to represent a metastatic lesion. Further evaluation with MRI revealed a benign vertebral hemangioma, highlighting the limitations of this radiotracer in the setting of benign spinal lesions. Multimodality imaging findings of metastatic MCC and potential pitfalls of Ga-68-DOTATATE PET-CT staging are discussed.

Foreign Body Aspiration Mimicking an Endobronchial Neoplasm: A Case Report and Review of the Literature.

Foreign body aspiration (FBA) is infrequently encountered in the adult population, with major risk factors including advancing age, intoxication, and disorders of the central nervous system. Here, we present a case of FBA in an adult undergoing routine lung cancer screening to review imaging findings and highlight potential pitfalls for the practicing radiologist. A low-dose chest computed tomography (CT) scan was performed for lung cancer screening in a 57-year-old male with a one-month history of worsening dyspnea and cough. An endobronchial lesion was identified in the right bronchus intermedius. A follow-up 18F-fluorodeoxyglucose positron emission tomography-computed tomography (18F-FDG PET-CT) revealed hypermetabolic activity in the region of interest, raising concern for malignancy. Bronchoscopy was performed, revealing a nodular mass adjacent to a foreign body in the bronchus intermedius. Histopathologic analysis of the tissue sample revealed the presence of an aspirated foreign body with squamous metaplasia of the respiratory epithelium. Adult FBA is an uncommon clinical entity that may be incidentally observed on a screening chest CT. Relevant multimodality imaging findings are discussed here, along with a review of the accompanying pathologic changes seen with chronic airway impaction.

Emphysematous Osteomyelitis: Novel Use of PET-MRI and a Review of Characteristic Imaging Findings.

INTRODUCTION: Emphysematous Osteomyelitis (EO) is an extremely rare bone infection caused by gas-forming bacteria with few documented cases in the literature. Our study aims to highlight characteristic imaging features, including the novel use of positron emission tomographymagnetic resonance imaging (PET-MRI) in diagnosing this potentially fatal entity. CASE: Radiography and computed tomography (CT) of the pelvis were performed due to complaints of persistent back pain in a 36-year-old male with a history of recent abdominal aorta surgery. Sacroiliac joint aspiration was performed, and a follow-up PET-MRI was subsequently performed. RESULTS: Radiography and CT demonstrated bilateral sacroiliitis, osteonecrosis and EO in the bony pelvis. Left sacroiliac joint aspiration identified Staphylococcus aureus as the causative organism. PET-MRI revealed EO with left iliopsoas abscess and abdominal aortic graft infection. The patient's symptoms resolved following antibiotic therapy and image-guided abscess drainage. CONCLUSION: EO is a lethal variant of osteomyelitis with a dearth of published cases. Pertinent imaging characteristics of EO on radiography, CT and PET-MRI are discussed here, along with a review of the literature surrounding this rare condition.

Pharmacogenetic Manipulation of the Nucleus Accumbens Alters Binge-Like Alcohol Drinking in Mice.

BACKGROUND: Chronic alcohol intake leads to long-lasting changes in reward- and stress-related neuronal circuitry. The nucleus accumbens (NAc) is an integral component of this circuitry. Here, we investigate the effects of DREADDs (Designer Receptors Exclusively Activated by Designer Drugs) on neuronal activity in the NAc and binge-like drinking. METHODS: C57BL/6J mice were stereotaxically injected with AAV2 hSyn-HA hM3Dq, -hM4Di, or -eGFP bilaterally into NAc [core + shell, core or shell]. We measured clozapine-n-oxide (CNO)-induced changes in NAc activity and assessed binge-like ethanol (EtOH) or tastant/fluid intake in a limited access Drinking in the Dark (DID) schedule. RESULTS: We found that CNO increased NAc firing in hM3Dq positive cells and decreased firing in hM4Di cells, confirming the efficacy of these channels to alter neuronal activity both spatially and temporally. Increasing NAc core + shell activity decreased binge-like drinking without altering intake of other tastants. Increasing activity specifically in the NAc core reduced binge-like drinking, and decreasing activity in the NAc core increased drinking. Manipulation of NAc shell activity did not alter DID. Thus, we find that increasing activity in the entire NAc, or just the NAc core is sufficient to decrease binge drinking. CONCLUSIONS: We conclude that the reduction in EtOH drinking is not due to general malaise, altered perception of taste, or reduced calorie-seeking. Furthermore, we provide the first evidence for bidirectional control of NAc core and binge-like drinking. These findings could have promising implications for treatment.

NPAS2 Regulation of Anxiety-Like Behavior and GABAA Receptors.

Abnormal circadian rhythms and circadian genes are strongly associated with several psychiatric disorders. Neuronal PAS Domain Protein 2 (NPAS2) is a core component of the molecular clock that acts as a transcription factor and is highly expressed in reward- and stress-related brain regions such as the striatum. However, the mechanism by which NPAS2 is involved in mood-related behaviors is still unclear. We measured anxiety-like behaviors in mice with a global null mutation in Npas2 (Npas2 null mutant mice) and found that Npas2 null mutant mice exhibit less anxiety-like behavior than their wild-type (WT) littermates (in elevated plus maze, light/dark box and open field assay). We assessed the effects of acute or chronic stress on striatal Npas2 expression, and found that both stressors increased levels of Npas2. Moreover, knockdown of Npas2 in the ventral striatum resulted in a similar reduction of anxiety-like behaviors as seen in the Npas2 null mutant mouse. Additionally, we identified Gabra genes as transcriptional targets of NPAS2, found that Npas2 null mutant mice exhibit reduced sensitivity to the GABAa positive allosteric modulator, diazepam and that knockdown of Npas2 reduced Gabra1 expression and response to diazepam in the ventral striatum. These results: (1) implicate Npas2 in the response to stress and the development of anxiety; and (2) provide functional evidence for the regulation of GABAergic neurotransmission by NPAS2 in the ventral striatum.

Functional Implications of the CLOCK 3111T/C Single-Nucleotide Polymorphism.

Circadian rhythm disruptions are prominently associated with bipolar disorder (BD). Circadian rhythms are regulated by the molecular clock, a family of proteins that function together in a transcriptional-translational feedback loop. The CLOCK protein is a key transcription factor of this feedback loop, and previous studies have found that manipulations of the Clock gene are sufficient to produce manic-like behavior in mice (1). The CLOCK 3111T/C single-nucleotide polymorphism (SNP; rs1801260) is a genetic variation of the human CLOCK gene that is significantly associated with increased frequency of manic episodes in BD patients (2). The 3111T/C SNP is located in the 3'-untranslated region of the CLOCK gene. In this study, we sought to examine the functional implications of the human CLOCK 3111T/C SNP by transfecting a mammalian cell line (mouse embryonic fibroblasts isolated from Clock(-/-) knockout mice) with pcDNA plasmids containing the human CLOCK gene with either the T or C SNP at position 3111. We then measured circadian gene expression over a 24-h time period. We found that the CLOCK3111C SNP resulted in higher mRNA levels than the CLOCK 3111T SNP. Furthermore, we found that Per2, a transcriptional target of CLOCK, was also more highly expressed with CLOCK 3111C expression, indicating that the 3'-UTR SNP affects the expression, function, and stability of CLOCK mRNA.