Artificial Intelligence's Prospects and pitfalls in the field of Diagnostic Radiology
| dc.creator | Verdier, Gavin | en_US |
| dc.creator.orcid | 0009-0008-9143-9858 (Verdier, Gavin) | |
| dc.date.accessioned | 2024-04-17T17:42:13Z | |
| dc.date.available | 2024-04-17T17:42:13Z | |
| dc.date.issued | 2024-03-21 | en_US |
| dc.description.abstract | Opening: Artificial Intelligence (AI) is a burgeoning field that is integrating into medicine at an accelerating pace. Its capacity for providing quantitative and qualitative image analysis is becoming an increasingly valuable tool for Diagnostic Radiology. Nowadays, a wide array of AI-based solutions are available, employing diverse approaches like Convolutional Neural Networks (CNNs), Reinforcement Learning, and Image and Texual Generation. Support: An area of implementation that is particularly intriguing is in the recent advancements in the interpretation of thoracic imaging. The interpretation of thoracic X-Rays in trauma settings might be delayed due to various hospital processes. Preliminary reads by AI algorithms could identify findings that could lower the workflow burden and allow physicians to focus on more challenging cases (Feng et al.). Additionally, datasets like those collected in these studies have contributed tens of thousands of high-quality, annotated X-Rays in open datasets, which can be used to continually enhance AI tools for the field. The AI technique of reconstruction can fill in the gaps of lower-quality imaging data, which has the potential to lower the amount of time and resources needed for scans such as Knee MRIs. These techniques could help to democratize advanced imaging to more patients and reduce facility operational costs. AI has the potential to provide standardization to the subjective nature of image interpretation, with one example being in evaluating metastatic breast lesions. However, Meta-analyses evaluating the efficacy of using AI in diagnosis and staging still display issues with sensitivity, with reported levels being as high as 20%, a figure unusable in a clinical setting. While approaches such as Deep Learning and Image Reconstruction have a promising future, the truth is that these technologies are still in their infancy. Datasets for Neuroimaging in MRIs might not contain sufficient data to provide the training necessary for certain reconstruction techniques, which could miss crucial artifacts in certain circumstances. There is a genuine concern as well that AI might exacerbate disparities in healthcare because of the inherent difference in how AI and human doctors arrive at decision-making. Concern arises around the interpretation of data and the potential for AI to propagate biases into the diagnostic image interpretation due to suboptimal AI training. Another example is that AI has been shown to deduce a patient's race without the relevant information as perceived by human clinicians being explicitly provided. The relative absence of understanding of the inner workings of AI algorithms can undermine the trustworthiness of many currently available AI systems. This, in turn, reflects the many potential areas in improving the interpretability of these often what regarded as “blackbox” systems. Conclusion: Though AI has many hurdles to overcome, it is likely to gain increasing importance in the Radiologist's toolkit over the coming years. These tools will necessitate the Radiologist to become even more familiar with the technology and data collection. The algorithms that are being developed will undoubtedly benefit from close collaboration between the physician, software developers, and engineers in order to develop effective and clinically relevant software. | en_US |
| dc.identifier.uri | https://hdl.handle.net/20.500.12503/32725 | |
| dc.language.iso | en | |
| dc.title | Artificial Intelligence's Prospects and pitfalls in the field of Diagnostic Radiology | en_US |
| dc.type | poster | en_US |
| dc.type.material | text | en_US |