Artificial Intelligence in Lymphoma Histopathology: Systematic Review

Area of Science:

• Medical Informatics
• Computational Pathology
• Oncology

Background:

• Artificial intelligence (AI) demonstrates potential in lymphoma diagnosis, prognosis, and gene prediction.
• A critical need exists for assessing AI model biases and clinical utility in lymphoma research.

Purpose of the Study:

• To systematically evaluate biases in published AI models for lymphoma histopathology.
• To assess the clinical utility of comprehensive AI models for lymphoma diagnosis and prognosis.

Main Methods:

• A systematic literature review followed PRISMA 2020 guidelines, searching PubMed, Cochrane Library, and Web of Science.
• The Prediction Model Risk of Bias Assessment Tool (PROBAST) was used to evaluate bias in 41 AI models.
• Data on AI models for lymphoma diagnosis, prognosis, and gene prediction were systematically tabulated.

Main Results:

• All 41 included AI models exhibited a high or unclear risk of bias, mainly due to incomplete reporting of participant selection and statistical analysis.
• Internal validation showed Area Under the Curve (AUC) from 0.75-0.99; external validation AUC ranged from 0.93-0.99.
• Commonly studied lymphoma subtypes included diffuse large B-cell lymphoma, follicular lymphoma, and chronic lymphocytic leukemia.

Conclusions:

• Methodological biases are prevalent in current AI models for lymphoma, impacting their clinical translation.
• Enhancing AI accuracy and clinical utility requires comprehensive reporting, diverse datasets, transparent models, and rigorous validation.