“Our findings confirm the robustness of AI models in predicting the clinical outcome of patients with lymphoma based on PET images,” noted lead meta-analysis author Mohammad Mehdi Mehrabi Nejad, M.D., who is affiliated with the Department of Radiology and the Advanced Diagnostic and Interventional Radiology Research Center (ADIR) at the Tehran University of Medical Sciences in Tehran, Iran, and colleagues.
“In the realm of precision medicine, AI methods can serve as a decision-support tool to physicians by providing valuable insights regarding an individual patient’s prognosis and further tailoring treatment regimens based on the patient’s unique characteristics. Leveraging these novel methods can result in improved patient outcomes by optimizing treatment regimens through patient-based risk stratification systems.”
Three Key Takeaways
1. AI-enhanced PET imaging significantly improves outcome prediction. Adjunctive use of AI in analyzing pre-treatment PET scans greatly increases the ability to predict progression-free survival (PFS) (over 4x) and overall survival (OS) (over 3x) in lymphoma patients.
2. Deep learning models outperform other AI approaches. Among the AI methods assessed, deep learning achieved the highest accuracy for predicting 2-year PFS (81 percent AUC), outperforming radiomic models and combined ML/radiomic models.
3. AI supports personalized treatment planning. AI tools can assist clinicians by enabling patient-specific risk stratification and tailoring of treatment strategies, potentially leading to improved outcomes in patients with lymphoma.
For the prediction of two-year PFS, the meta-analysis authors noted a pooled 75 percent AUC across 10 studies without any significant differences with respect to the type of lymphoma. The researchers pointed out that deep learning models offered the highest AUC for two-year PFS (81 percent) in contrast to 75 percent for the combination of machine learning and radiomics, and 73.3 percent for radiomic models.
“We found that among different AI methods, DL showed superior performance in predicting PFS, 2-year PFS, and treatment response based on PET imaging,” added Nejad and colleagues.
(Editor’s note: For related content, see “SNMMI: AI May Enhance Detection and Risk Assessment for Multiple Cancers on Whole-Body PET/CT Scans,” “Could Lymph Node Distribution Patterns on CT Improve Staging for Colon Cancer?” and “Study Examines Prognostic Value of Baseline PSMA PET/CT Factors in Patients with mCRPC.”)
In regard to limitations with the meta-analysis, the authors conceded high variability with respect to the types of AI, treatment protocols and lymphoma subtypes assessed in the reviewed studies. They also noted a limited number of studies on deep learning applications for patients with lymphoma and the exclusion of non-English publications.
