Application of artificial intelligence chatbots in interpreting magnetic resonance imaging reports: a comparative...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Medical Communication

Background:

• Interpreting complex radiological reports poses a challenge for patient comprehension.
• Artificial intelligence (AI) chatbots show potential for improving medical communication.
• The efficacy of AI chatbots in translating magnetic resonance imaging (MRI) reports is not well-established.

Purpose of the Study:

• To evaluate the performance of AI chatbots in translating MRI reports into patient-friendly language.
• To assess the accuracy of AI chatbots in classifying tumor characteristics and recommending treatments.
• To compare the readability and clinical relevance of AI-generated interpretations.

Main Methods:

• A cross-sectional analysis of 6174 MRI reports from tumor patients was conducted.
• Two AI chatbots, GPT o1-preview and Deepseek-R1, interpreted reports, classified tumors, assessed surgical necessity, and suggested treatments.
• Readability was measured using Flesch-Kincaid and Gunning Fog metrics; accuracy was verified by medical reviewers.

Main Results:

• Both chatbots significantly improved report readability, with Deepseek-R1 achieving higher Flesch-Kincaid Reading Ease scores.
• Deepseek-R1 demonstrated superior accuracy in tumor classification (92.05%) and surgical necessity assessment (95.12%).
• Deepseek-R1 provided more clinically relevant treatment recommendations (98.10% acceptable) and both showed high empathy.

Conclusions:

• AI chatbots, particularly Deepseek-R1, can effectively enhance MRI report readability and interpretation accuracy for patients.
• Despite high performance, AI chatbots are prone to errors like misinterpreting terminology and occasional hallucinations.
• Physician oversight is essential for safe clinical integration of AI chatbots in radiology, despite their potential to reduce healthcare burdens.