AI Simplifies CT Reports for Cancer Patients

A team at the Technical University of Munich (TUM) investigated how artificial intelligence can improve patient understanding of complex CT findings. Utilizing an open-source large language model operated on TUM University Hospital’s computers, the researchers simplified original reports for patients undergoing cancer diagnosis imaging. Results demonstrated a reduction in reading time from an average of seven minutes to two minutes, with patients reporting significantly improved comprehension—81% found the simplified texts easier to read compared to 17% with the original reports. This study highlights the potential for AI-supported simplification to enhance health literacy and informed consent.

AI Simplifies CT Reports for Patients

Artificial intelligence is demonstrating a benefit for cancer patients by simplifying complex CT reports. Researchers at the Technical University of Munich (TUM) investigated how AI could improve patient understanding of these reports, traditionally filled with specialized medical terminology. The study involved 200 patients, with one group receiving standard reports and the other receiving AI-simplified versions. Ensuring patient comprehension is considered a central pillar of modern medicine, vital for informed consent and health literacy.

The results showed a significant reduction in reading time, falling from an average of seven minutes for original reports to just two minutes with the AI-simplified texts. Patient satisfaction was also notably higher: 81% found the simplified reports easier to read compared to 17% with the originals. Furthermore, 80% found them easier to understand (compared to 9%), and 82% rated them as helpful and informative – far exceeding the 29% and 27% ratings for the standard reports.

While AI shows promise, the research team emphasizes the need for verification by health professionals. A review of the AI-generated reports revealed inaccuracies in 6% of cases, omissions in 7%, and the addition of new information in 3%. Therefore, the team cautions against using general chatbots like ChatGPT and stresses that language models are tools supporting medical staff, not replacements for them, to ensure patients receive accurate information.

Study Results: Reduced Reading Time & Increased Comprehension

A study by the Technical University of Munich investigated how artificial intelligence could improve patient understanding of CT reports. Results demonstrated a significant reduction in reading time, falling from an average of seven minutes for original reports to just two minutes with AI-simplified versions. Researchers included 200 patients with cancer diagnoses, dividing them to receive either standard or simplified reports, and confirmed improved readability through objective measurements alongside patient feedback.

Patients receiving the AI-simplified reports rated them as much easier to read (81% vs. 17%) and understand (80% vs. 9%), as well as more helpful (82% vs. 29%) and informative (82% vs. 27%) compared to those receiving the original reports. This indicates that AI-supported simplification positively impacts comprehension. The study, published in Radiology, suggests providing automatically simplified reports could be a beneficial service alongside specialist reports.

While the AI showed promise, researchers emphasize the need for review by health professionals. The investigation revealed inaccuracies in 6% of AI-generated findings, omissions in 7%, and additions of new information in 3%. Therefore, language models are viewed as useful tools but not substitutes for trained medical staff to ensure patients receive correct information about their illness.

AI Limitations & the Need for Professional Review

The study at the Technical University of Munich revealed limitations when using AI to simplify complex medical reports. Researchers found that 6% of AI-generated findings contained factual inaccuracies, while 7% omitted crucial information and 3% added new, potentially misleading, details. This underscores the necessity of professional oversight; reports were reviewed and corrected before being given to patients. The team emphasizes AI is a tool, not a replacement for trained medical staff who can verify accuracy.

The research demonstrated significant improvements in patient comprehension with AI-simplified CT reports. Reading time decreased from an average of seven minutes to just two minutes, and patient ratings for readability, understanding, helpfulness, and informativeness all increased substantially – from 17% to 81%, 9% to 80%, 29% to 82%, and 27% to 82% respectively. However, these benefits require a careful approach to implementation.

Researchers advocate for providing automatically simplified reports alongside specialist reports, but only with optimized, secure AI solutions and, crucially, professional verification. Dr. Philipp Prucker specifically advises against using chatbots like ChatGPT for simplification due to data protection concerns and the risk of factual errors. Language models, while useful, cannot substitute the expertise of medical staff in ensuring patients receive correct information about their illness.