Abstract
Purpose
This poster aims to review current literature on artificial intelligence (AI) in diagnostic imaging, relating this to human visual processes.
Advances in computational power have enabled new types of AI and machine learning with particular accuracy in image recognition. The functionality of these systems is not well understood due to the system architecture having hidden layers playing a large part in the decision-making process. This means that when errors are made, they are unexplainable to humans. This attributes to lack of trust in these systems. Work is currently being undertaken to clarify these processes and thus explain errors.
This poster aims to compare human and computer vision processes and explain these systems in an understandable way.
Methods
Data has been gathered from a wide range of sources, from peer-reviewed journals to AI builders’ information providing current perspective.
Results
Literature suggests that computational systems are being developed and tested which have progressed further than merely carrying out programmed function and are moving towards systems more in line with intuitive human visual systems. Results from this review indicate that there is still significant progress to be made to clarify the processes of machine learning systems and explain errors and are not fully relatable to human visual systems.
Conclusion
Significant advances have been made in recent years in computer vision, modelled on human visual perception. Differences prevail, with advantages in both computer and human visual systems. Promising outcomes in accuracy in radiographic image interpretation may exist when human and computer work synergistically.
This poster aims to review current literature on artificial intelligence (AI) in diagnostic imaging, relating this to human visual processes.
Advances in computational power have enabled new types of AI and machine learning with particular accuracy in image recognition. The functionality of these systems is not well understood due to the system architecture having hidden layers playing a large part in the decision-making process. This means that when errors are made, they are unexplainable to humans. This attributes to lack of trust in these systems. Work is currently being undertaken to clarify these processes and thus explain errors.
This poster aims to compare human and computer vision processes and explain these systems in an understandable way.
Methods
Data has been gathered from a wide range of sources, from peer-reviewed journals to AI builders’ information providing current perspective.
Results
Literature suggests that computational systems are being developed and tested which have progressed further than merely carrying out programmed function and are moving towards systems more in line with intuitive human visual systems. Results from this review indicate that there is still significant progress to be made to clarify the processes of machine learning systems and explain errors and are not fully relatable to human visual systems.
Conclusion
Significant advances have been made in recent years in computer vision, modelled on human visual perception. Differences prevail, with advantages in both computer and human visual systems. Promising outcomes in accuracy in radiographic image interpretation may exist when human and computer work synergistically.
| Original language | English |
|---|---|
| Publication status | Published (in print/issue) - 20 Jul 2021 |
| Event | ISRRT 2021 - Dublin/online, Ireland Duration: 20 Aug 2021 → 22 Aug 2021 https://isrrtdublin2021.org/ |
Conference
| Conference | ISRRT 2021 |
|---|---|
| Country/Territory | Ireland |
| Period | 20/08/21 → 22/08/21 |
| Internet address |
