Quote: The Data Quality Challenge in AI
An expert on a recent *Vizion Expert Series* podcast highlighted the primary obstacle to implementing AI in oncology: "90% of the time we spend just cleaning up data." The discussion emphasized that building canonical data models and ensuring high-quality, structured data are prerequisites for leveraging AI to reduce care variation and improve clinical decision support in complex fields like prostate cancer imaging.
- A primary challenge in developing effective AI models is the need for large volumes of high-quality, diverse, and standardized data, which is often difficult to gather from varied clinical settings. Inconsistent image formats, varying resolutions from different scanners, and incomplete patient metadata are common data quality issues that hinder AI implementation. - The market for AI in medical imaging is projected to grow significantly, with one forecast predicting a surge to over $20.40 billion by 2029, up from $5.86 billion in 2024. As of late 2025, the FDA had authorized 1,039 AI-enabled radiology devices, representing 77% of all such medical device approvals. - A major trend in imaging is the shift of services from hospitals to outpatient settings, with about 40% of all radiology volume now occurring in outpatient centers or clinics. Studies suggest that shifting even a fraction of hospital-based imaging to outpatient centers could lead to significant cost savings for both the healthcare system and patients. - This site-of-care shift is influenced by Medicare reimbursement policies, which have seen reductions for many imaging services. For instance, the 2025 Medicare Physician Fee Schedule includes a 2.83% reduction in the conversion factor, impacting reimbursement for many radiology procedures. - Private equity investment in diagnostic imaging is active, with firms acquiring 151 radiology practices between 2013 and 2023. This consolidation is driven by the potential for increased operational efficiency and greater negotiating power for reimbursement rates. - Radiology is facing a significant workforce shortage, with rising imaging volumes and radiologist attrition rates that have increased by 50% since 2020. This has led to increased workloads, with the busiest 25% of radiologists reading 31% more studies since 2018. - AI tools are being developed to improve workflow efficiency, with some studies showing the potential to reduce reading workflow times by as much as 79%. In one clinical study, a generative AI system boosted radiologist productivity by up to 40% without compromising accuracy. - Structured reporting is crucial for leveraging AI in radiology, as it allows algorithms to more effectively extract relevant features compared to unstructured text reports. This organized data is essential for training reliable machine learning models for tasks like classification and prediction.
