AI Shows Promise and Limitations Across Multiple Cancer Care Applications
Key Insights
AI demonstrates significant potential in cancer (search) diagnosis, with deep learning models achieving up to 94% accuracy in lung cancer prediction and substantially improving radiologist performance in image analysis.
Recent studies show AI language models reaching 85% accuracy on medical oncology exam questions, though concerns remain about potential clinical harm from incorrect responses.
AI applications are expanding into clinical trial design, patient recruitment, and treatment optimization, with ongoing trials evaluating AI platforms for immunotherapy and chemotherapy dose management.
Artificial intelligence is making significant inroads across multiple areas of oncology, with early diagnosis and imaging analysis emerging as particularly promising applications. However, experts emphasize the need for careful implementation and human oversight as the technology continues to evolve.
Leading oncology experts highlight several key areas where AI is showing immediate impact. "Community oncologists should understand that the applications of AI in oncology are going to be very broad," says Dr. Krishnansu S. Tewari, Director of the Division of Gynecologic Oncology at the University of California, Irvine School of Medicine (search). "There are going to be applications in clinical trial design, treatment response, triaging, imaging, pathology, genomics, transcriptomics, proteomics, [and] metabolomics."
AI Performance in Clinical Decision Support
Recent research has demonstrated both the potential and limitations of AI in clinical oncology. A cross-sectional study of large language models (LLMs) showed an impressive 85% accuracy rate on medical oncology examination questions. However, researchers found that incorrect answers could potentially cause moderate to severe harm in clinical practice, highlighting the importance of human oversight.
In imaging applications, AI is showing particularly strong results. A PET/CT-based deep learning radiomics model achieved 91% accuracy in predicting PD-L1 (search) expression in non-small cell lung cancer (search) patients. Even more impressive, an AI model named Sybil demonstrated accuracy rates between 86% and 94% in predicting future lung cancer (search) risk from single low-dose CT scans across multiple independent datasets.
Early Diagnosis and Imaging Analysis
"Probably the most powerful [current application of AI is for] early diagnosis," notes Kimberly Futch, MBA, Director of Clinical Operations Strategy at ProPharma (search). The technology is particularly effective when combined with human expertise, helping radiologists focus on critical findings and improving diagnostic accuracy.
A recent study examining 140 radiologists using AI across chest x-ray diagnostic tasks found significant improvements in accuracy, though with substantial variation in effectiveness among different practitioners.
Clinical Trial Applications and Challenges
AI is beginning to play a role in clinical trial design and patient recruitment. Two notable ongoing trials are evaluating the CURATE.AI platform for optimizing immunotherapy doses in solid tumors and chemotherapy doses in multiple myeloma (search) patients.
However, experts caution about limitations. "Before we roll AI out into clinical trial designs, [an area] where I do believe it will have an impact, we need to do so cautiously," Dr. Tewari emphasizes. "AI was developed for applications outside of the world of medicine and was not designed for medicine."
Current Limitations and Future Considerations
Several key challenges remain in AI implementation:
- Data Privacy: Ensuring patient confidentiality and proper data anonymization
- System Reliability: Maintaining accuracy across larger datasets
- Cost Considerations: Managing implementation expenses
- Bias and Consistency: Addressing potential algorithmic biases
- Pathophysiology Understanding: Recognizing AI's limitations in understanding disease mechanisms
"The most important shortcoming to be aware of with AI in oncology, particularly in clinical trials and drug development, is that AI has been loaded with illness scripts but does not understand pathophysiology," Dr. Tewari notes. "Therefore, human intervention and oversight are essential."
Despite these challenges, the consensus among experts is that AI will continue to enhance cancer (search) care as a supplementary tool. As Futch concludes, "This is all supplemental to what we have. It builds a stronger foundation for making good decisions of how to treat patients most effectively."