The role of artificial intelligence in radiology: decision-making, limitations, and human-AI collaboration
Abstract
Artificial intelligence (AI) has become one of the most actively discussed tools in modern day radiology, promising to help interpret X-rays, CT scans, and MRIs alongside human radiologists. It has matched or exceeded human accuracy outright, particularly in a few narrow tasks. Furthermore, AI has come at a time when imaging volume has grown far faster than the radiology workforce, leading to heavy workloads for radiologists. However, this progress is not without its issues, including AI models that can be deliberately fooled and an inability to explain how a given prediction was reached in the first place. What remains unresolved is whether this radiology-based AI can be reliably explained, trusted, and applied in a workplace setting when limitations such as the risk of human over-reliance and the availability of training data are considered. To address this, this paper examines peer-reviewed studies covering AI's technical validation, real-world clinical deployment, and behavioral evaluation of how radiologists work alongside this tool. The findings show that 1) a convolutional neural network can exceed average radiologist accuracy on a narrow pneumonia- detection task, 2) a structured radiologist-feedback system can significantly reduce a deployed model's false-positive rate, and 3) a study involving radiologists found AI's accuracy, not a radiologist's intelligence or experience, to be the strongest predictor of whether AI will help or hinder the user. Finally, this paper applies the knowledge from the results to a real-world AI versus radiologist comparison of a chest X-ray case, which results in a single page web interface development (https://yashcoder99.github.io/xray) via JavaScript. Together, this study demonstrates the complexity of AI's value in radiology. As AI usage and deployment rates accelerate and new peer-reviewed evidence becomes available every year, we offer an up-to-date, evidence-backed analysis and application for weighing AI's strengths against its somewhat hidden weaknesses.