Best ChatGPT prompts for Radiologists

Given a patient with a history of claustrophobia and a suspected spinal cord injury, generate a customized MRI protocol that prioritizes image quality while minimizing scan time and patient discomfort. Consider the following parameters: 3T scanner, spine coil, axial and sagittal sequences, and a maximum scan time of 20 minutes. Provide a detailed sequence of scans, including slice thickness, spacing, and orientation, as well as recommendations for patient positioning and breathing instructions.

Analyze the DICOM headers of a set of CT scans from a multi-center clinical trial to identify inconsistencies in patient demographics, scan protocols, and image acquisition parameters. Compare the header information against a predefined set of standards and generate a report highlighting any discrepancies, along with recommendations for data correction and standardization. Assume the scans are stored in a PACS system and provide a step-by-step guide for extracting and analyzing the header information using a programming language like Python.

Develop a quality audit checklist for radiology reports based on the ACR guidelines, focusing on key elements such as clinical history, comparison to prior studies, and recommendation for further imaging or follow-up. Apply this checklist to a sample set of 20 radiology reports from a recent quality improvement project, and generate a summary report highlighting areas of excellence and opportunities for improvement. Include metrics on report completeness, accuracy, and consistency, as well as suggestions for report template modifications and radiologist training.

Validate the performance of a deep learning-based image segmentation algorithm for detecting lung nodules in chest CT scans, using a publicly available dataset like LIDC-IDRI. Compare the algorithm's output against manual segmentations performed by experienced radiologists, and calculate metrics such as Dice similarity coefficient, precision, and recall. Provide a detailed analysis of the algorithm's strengths and weaknesses, including any bias towards specific nodule types or sizes, and suggest potential improvements to the algorithm's architecture or training data.