Artificial Intelligence in Medical Diagnosis

 Artificial Intelligence in Medical Diagnosis 

            In today’s world, we see AI everywhere, from our very own phones to advanced technology in detecting medical conditions even before the real professionals can. As artificial intelligence becomes more involved in areas of expertise such as healthcare, we are seeing how these technological advances are more important than ever. The research conducted at Loyola University on “Beyond the Contour: How 3D cues enhance object recognition in human and neural networks” by Cutler and Baumel et al. studied how the human and artificial neural networks or ANNs use 3D cues to recognize objects (Cutler & Baumel, 2025). Similarly, the recent study from Stanford Human-Centered Artificial Intelligence (HAI) shows the improving accuracy in diagnostics in medical imaging with the help of AI. Both studies show the implications of AI in performance, while also highlighting the access to bigger and broader data sets and a need for deeper 3D and relational understanding, not just instrumental. 

            In the Stanford HAI study, they demonstrated that multimodal large language models (LLMs), being paired with image data, such as X-rays, can outperform specialists such as radiologists in tasks such as coming up with a diagnosis and with a lower number of errors and biases (Stanford HAI, 2024). This mimics how human clinicians combine the structural visual cues they get with contextual reasoning they have already acquired, which is similar to how, in the Cutler and Baumel study, humans outperformed ANNs when trying to recognize objects from noncanonical perspectives that used volumetric shading (Cutler & Baumel, 2025). This is telling us that human perception relies on many things that the ANNs still don’t have, such as relational processing. 

            Findings on both of these studies show us the human expertise, whether it be in object recognition or medical diagnosis, the use of artificial intelligence must be improved in a way that it is not only extracting information on a surface level but rather deeper knowledge that can support inference and reasoning behind it. Particularly in the Cutler and Baumel et al. study, the ANNs must be improved in order to learn to handle real-world data that may contain topics such as ambiguity and variability. The low capacity for flexibility and interpretation in technologies such as 3D and AI can be detrimental in cases where they deviate from their training, which is precisely where human insight is most needed. 

            The challenge that this new technology faces is not only about improvement, but also about ensuring that AI can generalize scenarios and implement ethics and diversity in such unpredictable scenarios. These systems, just like humans, can always improve and learn in order to achieve a level of sophistication and accuracy that can be trusted and reliable in the real world. 

            

 

References

Cutler, M., & Baumel, L. D., Tocco, J., Friebel, W., Thirivathukal, G. K., & Baker, N. (2025). Beyond the contour: How 3D cues enhance object recognition in humans and neural networks. Journal of Vision. (Pre-publication draft) 

Stanford HAI. (2024, April 15). Can AI improve medical diagnostic accuracy? Stanford Institute for Human-Centered Intelligence. https://hai.stanford.edu/news/can-ai-improve-medical-diagnostic-accuracy