Comparison of Ki67 Proliferation Index in Gastrointestinal Non-Hodgkin Large B-Cell Lymphomas: The Conventional Method of Evaluation or AI Evaluation?

Abstract

The most common NHL subtype in SEEU is DLBCL (39%), and it manifests with a variety of cellular morphologies and a high proliferation index. Also, the GI tract is the most common site of extranodal NHLs, and most NHLs involving the GI tract are of B-cell lineage, of which diffuse large B-cell lymphoma is the most common subtype, irrespective of location. The last few years have seen digital pathology as a vital technology that has a positive impact on diagnostics, but studies on the use of DP for lymphoma identification, however, are still restricted to only determining whether a tumor is present or absent. Using the example of cases of malignant NHL, we aim to investigate the diagnostic utility of DP using QuPath software in evaluating the proliferation index and the prognostic significance and to show that improved visualization and analysis contribute to the convergence of these complementary diagnostic modalities for lymphomas. The average proliferation index (Ki67) was 58.33% with values between 10% and 85%. After the stratification of the cases, an increased proliferation index was observed in the majority of cases (53.33%), and this aspect was associated with the advanced age of the patients (p = 0.045). Visual assessment provides lower Ki67 values than automated digital image analysis. However, the agreement coefficient between the conventional method and the AI method indicates an excellent level of reliability (ICC1–0.970, ICC2–0.990). The multivariate analysis revealed that in the cases where the proliferation index Ki67 is high (˃70%), the IPI score represents an important risk factor predicting mortality (HR = 10.597, p = 0.033).

Keywords

• artificial intelligence
• computer-assisted image interpretation
• pathology
• lymphoma
• large B cell
• diffuse