Does the new AI test work well for distinguishing Acute Promyelocytic Leukemia from other types?
Acute Promyelocytic Leukemia (APL) is a specific type of acute myeloid leukemia defined by a genetic fusion called PML::RARA. Standard tests like PCR can sometimes miss rare variations of this fusion, making diagnosis difficult. A new AI tool has been tested to help identify APL subtypes using standard lab results.
What the research says
In a study of 6,206 patients from 20 international centers, an AI algorithm predicted APL with an area under the curve (AUROC) of 0.98. This metric indicates very high accuracy for distinguishing this subtype from others like acute lymphoblastic leukemia (ALL) or standard acute myeloid leukemia (AML) 4. The tool also showed strong performance for AML overall with an AUROC of 0.94 4.
However, the initial version of this AI model had a significant limitation regarding patient coverage. When set to a high confidence level to ensure accuracy, the tool excluded between 70.8% and 92.5% of patients from its predictions 4. Researchers improved the model by using an ensemble of Isolation Forest and Local Outlier Factor methods. This adjustment increased accuracy for AML while reducing the number of excluded patients to only 12.1% 4.
While the AI tool performs well on paper, real-world diagnosis often involves complex cases where standard tests fail. For example, some patients have cryptic or atypical genetic fusions that escape detection by routine PCR testing 2. In such cases, RNA sequencing can identify these rare variants, and the AI tool may need to be validated against these advanced sequencing results to ensure it works well in all scenarios 2. Additionally, rare presentations like spinal cord compression can complicate the clinical picture, requiring careful interpretation alongside AI outputs 1.
What to ask your doctor
- How does your lab currently test for APL, and do they use RNA sequencing if standard PCR tests are negative?
- Would an AI-based diagnostic tool be available or useful for my specific lab results?
- What are the limitations of AI tools regarding patient exclusion rates in your current workflow?
- If my case involves rare genetic variations, how will the diagnosis be confirmed beyond standard tests?
This question is drawn from common patient questions about this topic and answered using cited medical research. We do not provide individualized advice.