Ramin Javan

Prompt engineering and fine-tuning are powerful tools that can maximize the potential of large language models for their use in radiology, yet many challenges and limitations remain for their full implementation.

The next step in the evolution of static 3-dimensionally (3D) printed models may be the creation of “smart” models, where subcomponents can be seamlessly interacted with through a feedback mechanism, with potential applications in trainee education and patient counseling. Considering the complexity of the ventricular and cisternal systems, they were chosen for segmentation, using Materialize InPrint with outward hollowing using 2.5-mm wall thickness. After 3D printing, dedicated holes were drilled for placement of wired light emitting diodes (LEDs) in anatomical landmarks and connected to an Arduino Uno microcontroller. This was coupled to a Bluetooth transceiver for communication via an Android cellular device. C++ was used to match each LED to a particular pin number on the Arduino board. When the user verbalizes a structure, the Bluetooth sends a command to the Arduino, where the code looks for the “trigger word,” subsequently sending a signal to illuminate the corresponding LED. The system requires wireless/cellular data for communication with the voice recognition engine on the Google server. The described method may serve as a prototype for when 3D printers are capable of simultaneously printing conductive material or wiring along with the main material within a model to allow for integration of feedback devices.

AI-powered multimedia generation technologies, particularly video synthesis through stable diffusion and transformers, offer transformative potential for radiology education, communication, and visualization. This study explores various AI-generated multimedia categories, including image and video generation, as well as voice cloning, with a focus on video synthesis and future possibilities like scan-to-video generation. Utilizing tools such as Midjourney, RunwaymL Gen2, D-ID, and ElevenLabs, we aimed to reincarnate deceased influential physicists in radiology, demonstrating AI's capability to generate realistic content with accessible tools, fostering creativity and innovation in the radiology community.

Background Recent advancements in artificial intelligence (AI)-powered image generation present opportunities to enhance three-dimensional medical images. Diffusion, an iterative denoising process, represents the standard of many of the current tools used for this purpose. Purpose To demonstrate the current capabilities of diffusion technology by using Midjourney, version 5.2, a text-to-image generative AI tool, and present a practical guide for its use. Materials and Methods This exploratory study investigates the principles, parameters, and prompt engineering techniques for generating images focusing on Midjourney from July 27 to August 3, 2023. Step-by-step instructions show the innate capability of this technology in creating realistic medical images. Results Thirty images were selected, including eye, skin, and vascular aneurysm images. Varying prompt phrasing and weighting techniques allowed for the customization of output image characteristics. Although the details of Midjourney's model training are confidential, it is estimated that it was trained on at least hundreds of millions of images from the web. Anatomic fidelity was not always maintained because the training data set is not necessarily based on accurate medical images. There are shortcomings in this nascent technology regarding its ability to create entities such as digits of the hand or precise text. Conclusion AI image generation has the potential to improve three-dimensional medical images for certain applications through added visual detail and appeal but ongoing collaboration is needed between radiologists and AI developers due to the overreliance on art and photography in the training data, which may result in inaccurate anatomic results. Moreover, the evolving landscape of ethical discussions and copyright stipulations warrants close attention. © RSNA, 2024 Supplemental material is available for this article.

ChatGPT, a large language model by OpenAI, has been adopted in various domains since its release in November 2022, but its application in ophthalmology remains less explored. This editorial assesses ChatGPT's potential applications and limitations in ophthalmology across clinical, educational, and research settings. In clinical settings, ChatGPT can serve as an assistant, offering diagnostic and therapeutic suggestions based on patient data and assisting in patient triage. However, its tendencies to generate inaccurate results and its inability to keep up with recent medical guidelines render it unsuitable for standalone clinical decision-making. Data security and compliance with the Health Insurance Portability and Accountability Act (HIPAA) also pose concerns, given ChatGPT’s potential to inadvertently expose sensitive patient information. In education, ChatGPT can generate practice questions, provide explanations, and create patient education materials. However, its performance in answering domain-specific questions is suboptimal. In research, ChatGPT can facilitate literature reviews, data analysis, manuscript development, and peer review, but issues of accuracy, bias, and ethics need careful consideration. Ultimately, ensuring accuracy, ethical integrity, and data privacy is essential when integrating ChatGPT into ophthalmology.

Computed tomography angiography (CTA) of the head and neck is central in emergency department (ED) evaluation of clinically suspected acute stroke and intracranial hemorrhage. Timely and accurate detection of acute findings is crucial for best clinical outcomes; missed or delayed diagnosis can be devastating. Our pictorial essay presents twelve CTA cases that provided significant diagnostic dilemmas to on-call trainees while reviewing current bias and error classifications in radiology. Among others, we discuss anchoring, automation, framing, satisfaction of search, scout neglect and zebra-retreat bias. Each imaging vignette depicts a potential diagnostic “pitfall” while introducing types of cognitive bias/error before concluding with a concrete “pearl” for CTA interpretation. We believe that familiarity with bias and error is particularly important in the ED setting where high case volume, high acuity and radiologist fatigue intersect. Particular attention to personal cognitive biases and these potential CTA pitfalls may help emergency radiologists transition from habit-driven pattern recognition to analytical thinking, ultimately improving diagnostic decision making.