Vision-Based ChatGPT Shows Deficits Interpreting Radiologic Images

Researchers evaluating the performance of ChatGPT-4 Vision found that the model performed well on text-based radiology exam questions but struggled to answer image-related questions accurately. The study's results were published today in Radiology, a journal of the Radiological Society of North America (RSNA).

Chat GPT-4 Vision is the first version of the large language model that can interpret both text and images.

"ChatGPT-4 has shown promise for assisting radiologists in tasks such as simplifying patient-facing radiology reports and identifying the appropriate protocol for imaging exams," said Chad Klochko, M.D., musculoskeletal radiologist and artificial intelligence (AI) researcher at Henry Ford Health in Detroit, Michigan. "With image processing capabilities, GPT-4 Vision allows for new potential applications in radiology."

For the study, Dr. Klochko’s research team used retired questions from the American College of Radiology’s Diagnostic Radiology In-Training Examinations, a series of tests used to benchmark the progress of radiology residents. After excluding duplicates, the researchers used 377 questions across 13 domains, including 195 questions that were text-only and 182 that contained an image.

GPT-4 Vision answered 246 of the 377 questions correctly, achieving an overall score of 65.3%. The model correctly answered 81.5% (159) of the 195 text-only queries and 47.8% (87) of the 182 questions with images.

"The 81.5% accuracy for text-only questions mirrors the performance of the model’s predecessor," he said. "This consistency on text-based questions may suggest that the model has a degree of textual understanding in radiology."

Genitourinary radiology was the only subspecialty for which GPT-4 Vision performed better on questions with images (67%, or 10 of 15) than text-only questions (57%, or 4 of 7). The model performed better on text-only questions in all other subspecialties.

The model performed best on image-based questions in the chest and genitourinary subspecialties, correctly answering 69% and 67% of the image-containing questions, respectively. The model performed lowest on image-containing questions in the nuclear medicine domain, correctly answering only 2 of 10 questions.

The study also evaluated the impact of various prompts on the performance of GPT-4 Vision.

  • Original: You are taking a radiology board exam. Images of the questions will be uploaded. Choose the correct answer for each question.
  • Basic: Choose the single best answer in the following retired radiology board exam question.
  • Short instruction: This is a retired radiology board exam question to gauge your medical knowledge. Choose the single best answer letter and do not provide any reasoning for your answer.
  • Long instruction: You are a board-certified diagnostic radiologist taking an examination. Evaluate each question carefully and if the question additionally contains an image, please evaluate the image carefully in order to answer the question. Your response must include a single best answer choice. Failure to provide an answer choice will count as incorrect.
  • Chain of thought: You are taking a retired board exam for research purposes. Given the provided image, think step by step for the provided question.

Although the model correctly answered 183 of 265 questions with a basic prompt, it declined to answer 120 questions, most of which contained an image.

"The phenomenon of declining to answer questions was something we hadn’t seen in our initial exploration of the model," Dr. Klochko said.

The short instruction prompt yielded the lowest accuracy (62.6%).

On text-based questions, chain-of-thought prompting outperformed long instruction by 6.1%, basic by 6.8%, and original prompting style by 8.9%. There was no evidence to suggest performance differences between any two prompts on image-based questions.

"Our study showed evidence of hallucinatory responses when interpreting image findings," Dr. Klochko said. "We noted an alarming tendency for the model to provide correct diagnoses based on incorrect image interpretations, which could have significant clinical implications."

Dr. Klochko said his study’s findings underscore the need for more specialized and rigorous evaluation methods to assess large language model performance in radiology tasks.

"Given the current challenges in accurately interpreting key radiologic images and the tendency for hallucinatory responses, the applicability of GPT-4 Vision in information-critical fields such as radiology is limited in its current state," he said.

Hayden N, Gilbert S, Poisson LM, Griffith B, Klochko C.
Performance of GPT-4 with Vision on Text- and Image-based ACR Diagnostic Radiology In-Training Examination Questions.
Radiology. 2024 Sep;312(3):e240153. doi: 10.1148/radiol.240153

Most Popular Now

Unlocking the 10 Year Health Plan

The government's plan for the NHS is a huge document. Jane Stephenson, chief executive of SPARK TSL, argues the key to unlocking its digital ambitions is to consider what it...

Alcidion Grows Top Talent in the UK, wit…

Alcidion has today announced the addition of three new appointments to their UK-based team, with one internal promotion and two external recruits. Dr Paul Deffley has been announced as the...

AI can Find Cancer Pathologists Miss

Men assessed as healthy after a pathologist analyses their tissue sample may still have an early form of prostate cancer. Using AI, researchers at Uppsala University have been able to...

AI, Full Automation could Expand Artific…

Automated insulin delivery (AID) systems such as the UVA Health-developed artificial pancreas could help more type 1 diabetes patients if the devices become fully automated, according to a new review...

How AI could Speed the Development of RN…

Using artificial intelligence (AI), MIT researchers have come up with a new way to design nanoparticles that can more efficiently deliver RNA vaccines and other types of RNA therapies. After training...

MIT Researchers Use Generative AI to Des…

With help from artificial intelligence, MIT researchers have designed novel antibiotics that can combat two hard-to-treat infections: drug-resistant Neisseria gonorrhoeae and multi-drug-resistant Staphylococcus aureus (MRSA). Using generative AI algorithms, the research...

AI Hybrid Strategy Improves Mammogram In…

A hybrid reading strategy for screening mammography, developed by Dutch researchers and deployed retrospectively to more than 40,000 exams, reduced radiologist workload by 38% without changing recall or cancer detection...

Penn Developed AI Tools and Datasets Hel…

Doctors treating kidney disease have long depended on trial-and-error to find the best therapies for individual patients. Now, new artificial intelligence (AI) tools developed by researchers in the Perelman School...

New Training Year Starts at Siemens Heal…

In September, 197 school graduates will start their vocational training or dual studies in Germany at Siemens Healthineers. 117 apprentices and 80 dual students will begin their careers at Siemens...

Are You Eligible for a Clinical Trial? C…

A new study in the academic journal Machine Learning: Health discovers that ChatGPT can accelerate patient screening for clinical trials, showing promise in reducing delays and improving trial success rates. Researchers...

New AI Tool Addresses Accuracy and Fairn…

A team of researchers at the Icahn School of Medicine at Mount Sinai has developed a new method to identify and reduce biases in datasets used to train machine-learning algorithms...

Global Study Reveals How Patients View M…

How physicians feel about artificial intelligence (AI) in medicine has been studied many times. But what do patients think? A team led by researchers at the Technical University of Munich...