Artificial Intelligence in Digital Radiography: A Systematic Review of Evidence, Limitations, and Research Gaps in Collimation Accuracy

Background: We are seeing a lot of talk about Artificial Intelligence (AI) taking over digital radiography, especially for things like making workflows smoother or cutting down radiation doses. But here is the problem: we still don't really know if AI can actually handle "Collimation Accuracy." Since collimation is what keeps an image sharp and a patient safe from unnecessary radiation, it's a huge blind spot in our current knowledge. Objective: In this review, we set out to see what the evidence actually says. We wanted to find out if AI is truly helping with collimation and dose reduction, or if it’s just being used for basic behind-the-scenes office work in clinics. Methods: We didn't just look anywhere; we did a deep dive into PubMed, Scopus, and Web of Science. Following the PRISMA 2020 path, we hunted for any study that used AI or Deep Learning in a digital X-ray setting. To make sure we weren't looking at "junk science," we used the JBI appraisal tools to check if the research was actually solid and trustworthy. Results: After starting with 170 papers, we were surprised to find that only 7 actually made the cut. Even more shocking? Only two of those studies even tried to put AI into a real-life clinical workflow. As for "Automated Collimation" based on a patient’s actual anatomy—there was absolutely nothing. Most of the claims about AI saving dose were indirect and, frankly, the study quality was mostly just low-to-moderate according to our JBI check. Conclusion: The bottom line is that AI looks good on paper for workflow, but when it comes to the actual "hands-on" job of collimating an X-ray, the evidence is missing. We have a massive gap here. We need real-world, validated AI models that can adjust the X-ray beam in real-time if we want to take patient safety to the next level.