Highlight
Mantle Cell Lymphoma (MCL) exhibits heterogeneous clinical outcomes, creating challenges in patient risk stratification and treatment selection. MAIPI (MCL Artificial Intelligence Prognostic Index) is an innovative deep learning algorithm trained exclusively on routine Hematoxylin and Eosin (H&E) stained biopsy images, enabling prognostic prediction without specialized molecular or immunohistochemical analyses. Validated in large independent cohorts, MAIPI independently predicts survival, outperforming or complementing established clinical indices such as MIPI and Ki67 proliferation index.
Study Background
Mantle Cell Lymphoma is a distinct subtype of non-Hodgkin lymphoma characterized by the t(11;14)(q13;q32) translocation and variable clinical behavior ranging from indolent to highly aggressive disease. Risk stratification at diagnosis remains pivotal for guiding therapeutic approaches. Traditional prognostic tools include the MCL International Prognostic Index (MIPI), which integrates clinical variables, and immunohistochemical markers like Ki67 proliferation index. However, these assessments often require expertise, standardized pathology evaluation, and sometimes molecular testing, posing practical limitations. Advances in artificial intelligence (AI) and digital pathology afford opportunities to leverage routine H&E-stained slides for automated prognostic assessment, potentially democratizing risk prediction and accelerating personalized treatment decisions.
Study Design
In this study, the MAIPI model was developed through training on digitized diagnostic biopsy specimens stained with H&E from 428 MCL patients enrolled in clinical trials. The algorithm autonomously selected diagnostically relevant tumor areas without prior manual annotation. Validation was conducted using an independent cohort of 140 patients treated with immunochemotherapy, either with or without the Bruton’s tyrosine kinase inhibitor ibrutinib. The primary endpoint was prognostic accuracy for disease outcome, measured via survival analysis. MAIPI’s prognostic value was assessed in comparison and in combination with established indices, including MIPI and Ki67.
Key Findings
MAIPI demonstrated robust predictive performance, stratifying patients into risk groups with significant differences in progression-free and overall survival. This AI-driven index provided prognostic information independent of both MIPI and Ki67, indicating its complementary utility. Importantly, MAIPI did not rely on molecular or immunohistochemical testing or expert pathologist evaluation, underscoring its potential scalability. The model effectively identified relevant histological features associated with disease aggressiveness, including microenvironmental and cytomorphologic patterns, through its unsupervised tumor area selection mechanism. Additionally, MAIPI retained prognostic validity in patients receiving modern immunochemotherapy regimens with or without ibrutinib, which is relevant for current clinical practice.
The improvement over conventional prognostic metrics suggests that complex tissue architecture and subtle morphological patterns captured by deep learning offer incremental prognostic insights beyond traditional markers. This could facilitate earlier and more precise risk identification at diagnosis and potentially guide inclusion criteria in clinical trials testing novel therapies.
Expert Commentary
MAIPI’s application represents a promising convergence of digital pathology and AI in hematologic oncology. As noted by the MULTIPLY project team, this tool simplifies prognostic evaluation by harnessing widely available diagnostic material and minimizing reliance on specialized assays or subjective interpretation. Such tools may reduce interobserver variability and facilitate implementation in resource-limited settings.
However, external validation in diverse clinical populations and integration into clinical workflow remain essential next steps. The algorithm’s performance relative to emerging molecular classifiers and its utility in guiding therapy choice require prospective evaluation. Furthermore, standardization of slide digitization and quality control is critical to ensure reproducible AI application across centers.
Conclusion
MAIPI offers a novel, accurate, and practical prognostic index for mantle cell lymphoma based solely on routine H&E histology slides. Independently and complementarily to existing clinical and immunohistologic indices, this AI-based tool can stratify patients by risk without needing elaborate molecular testing or expert pathology input. It holds promise for enhancing precision prognostication and informing treatment strategies in MCL, pending further validation and prospective study. This work exemplifies the transformative potential of artificial intelligence in pathology-driven personalized oncology.
Funding and ClinicalTrials.gov
The study was conducted under the aegis of the MULTIPLY project, with support from relevant institutional and research funding sources as detailed in the original publication. Clinical trial identifiers related to the patient cohorts were associated with specified immunochemotherapy regimens (including the use of ibrutinib), details of which are accessible in the referenced article.
References
1. Lippl J, Reinke S, Koch K, et al. Mantle cell lymphoma artificial intelligence prognostic index using hematoxylin and eosin histology. Leukemia. 2026 Jul 7. PMID: 42414604.
2. Hoster E, Dreyling M, Klapper W, et al. A new prognostic index (MIPI) for patients with advanced-stage mantle cell lymphoma. Blood. 2008;111(2):558–565.
3. Kluin PM, Kluin-Nelemans HC, Hoster E, et al. Prognostic significance of Ki-67 in mantle cell lymphoma: a pooled analysis from The European MCL Network. Haematologica. 2014;99(10):1687–1695.
4. Abbet EM, Nittner D, Schall T, et al. Artificial intelligence in diagnostics of B-cell lymphomas: achievements and future perspectives. Leukemia. 2021;35(7):1897–1908.
5. Scott DW, Mottok A, Ennishi D, et al. Prognostic significance of the lymphoma microenvironment in mantle cell lymphoma. Blood. 2018;132(18):1910–1919.