AI Tool Reads Biopsy Images to Predict Mantle Cell Lymphoma Outcomes

A deep learning tool trained on routine biopsy slides predicted how patients with mantle cell lymphoma fared, matching prognostic information that otherwise requires specialized Ki-67 staining or TP53 molecular testing, according to a study published in Leukemia.¹

Mantle cell lymphoma (MCL) is a rare, often aggressive subtype of non-Hodgkin lymphoma, and clinicians have long combined the MCL International Prognostic Index (MIPI) with Ki-67 proliferation staining or TP53 mutation testing to flag patients at the highest risk of early relapse or death. Ki-67 scoring requires a trained hematopathologist following strict guidelines, and TP53 testing depends on molecular sequencing that can be slow, costly, and unavailable in some treatment settings, the study authors noted.

The MIPI itself was developed from a cohort of patients with advanced-stage disease, combining age, performance status, lactate dehydrogenase level, and white blood cell count into a single risk score. Building on that framework with Ki-67 or TP53 data improves its accuracy, but only where that additional testing is accessible, leaving a gap the new study's authors set out to close with a tool built entirely from images already captured at diagnosis.²

AI Model Built From Biopsy Images in 2 Clinical Trial Cohorts

The present researchers trained their artificial intelligence tool, called the Mantle Cell Lymphoma Artificial Intelligence Prognostic Index (MAIPI), on hematoxylin and eosin-stained biopsy images from 428 patients in a development cohort, then tested it on an independent cohort of 140 patients.¹ Samples came from the MCL Younger and MCL Elderly trials, both registered between 2004 and 2014 with follow-up extending up to 16 years, and from the phase 3 TRIANGLE (NCT02858258) trial, which had a median follow-up of 4.66 years. In total, the analysis drew on 568 patients using 309 tissue microarray samples and 259 whole slide images.

Ki-67 results were available for 527 of those patients and TP53 status for 326, letting the researchers compare MAIPI directly against the established biomarkers in the same patients rather than relying on separate historical benchmarks.

MAIPI Matched Established Biomarkers for Predicting Survival

In the independent test cohort, MAIPI predicted overall survival with a concordance index (C-index) of 0.673 (P = .003) and time to progression with a C-index of 0.677 (P < .001). Patients in the highest-risk quartile by MAIPI score had significantly worse overall survival than those in the lowest-risk quartile in both the development cohort (P < .001) and the test cohort (P = .013).

Adding MAIPI to the standard MIPI score improved its predictive accuracy for overall survival, raising the C-index by 0.027 in the development cohort and 0.002 in the test cohort, and the tool remained statistically significant for both overall survival and time to progression even after adjusting for MIPI (P < .001 for each). Combining MAIPI with Ki-67 similarly preserved its prognostic value (P = .023 for overall survival; P = .039 for time to progression), and MAIPI scores were higher in the more aggressive pleomorphic and blastoid MCL subtypes than in classic or small cell variants.

“The MAIPI is potentially available at the time of diagnosis and highly significant for prognosis,” the researchers wrote.

Independent Validation and Non-Nodal Disease Remain Open Questions

The authors noted that MAIPI's prognostic value needs confirmation in cohorts beyond the trial populations studied here and that its performance on biopsies taken at relapse, rather than at initial diagnosis, remains untested. Follow-up in the TRIANGLE cohort was shorter and included fewer events than the older trials, and the researchers cautioned that MAIPI's prognostic power specifically in ibrutinib-treated patients will need longer follow-up or additional validation cohorts.

Very few patients with non-nodal MCL, a less common presentation of the disease, were included, so the authors could draw no firm conclusions for that subtype. For now, they characterized MAIPI as a tool that captures subtle morphologic features not visible to the eye, extending rather than replacing existing risk models.

For clinicians, an AI tool that extracts prognostic information directly from biopsy slides already collected at diagnosis could flag high-risk patients with mantle cell lymphoma in settings where Ki-67 expertise or TP53 sequencing is limited, potentially speeding decisions about treatment intensity. For payers and health systems, broader validation could support using this lower-cost, single-slide approach to triage patients toward more aggressive regimens or closer monitoring without adding specialized testing to the diagnostic workup.

References

1. Lippl J, Reinke S, Koch K, et al. Mantle cell lymphoma artificial intelligence prognostic index using hematoxylin and eosin histology. Leukemia. 2026;40(9):1961-1967. doi:10.1038/s41375-026-03036-8
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. doi:10.1182/blood-2007-06-095331