PROJECT SUMMARY

CoPI with Oscar Brück, E.M., M.D., Ph.D., Helsinki University Hospital and University of Helsinki

Background

Mutations in genes regulating blood cell development give a competitive advantage to altered cells and can affect blood cell counts. While these are critical early steps towards MDS and AML, there are neither affordable nor scalable strategies to identify these at the population scale and focus active monitoring or early treatment measures.

Clinical models to estimate the prognosis of MDS patients highlight blood cell counts and alterations in chromosomes and genes. Despite its irrefutable significance in the diagnosis of MDS, the role of white blood cell proportion and morphology to guide treatment decisions has been barely studied due to the lack of large datasets of blood cell images.

By applying advanced artificial intelligence (AI) algorithms, we have recently found that bone marrow tissue composition can “see” genetic alterations and prognostic biomarkers from images. However, similar studies have not been conducted in more common peripheral blood and bone marrow smear slides.

Hypothesis

We hypothesize that AI-driven image analysis of bone marrow and peripheral blood slides can predict future occurrence and disease outcome of MDS and AML patients.

Objective

The proposal has three major objectives

1. Image analysis of bone marrow pathology slides together with clinical data can identify gene alterations and predict the treatment response of MDS and AML patients.
2. Image analysis-based analysis of peripheral blood samples can reliably screen for future MDS and AML.
3. Validation of the Objective 2 algorithms in a prospective medical device trial.

The study is a transatlantic two-center collaboration between the Helsinki University Hospital (covering 2 million inhabitants) in Finland, and the Brigham and Women’s Hospital/Dana Farber Institute. For Objective (1), we will study a unique collection of 22,000 bone marrow samples from 2,000 MDS/AML patients. For Objectives (2) and (3), we will include any sample for which imaging-based blood cell analysis has been performed as part of routine blood cell count (80,000 samples in Helsinki and 100,000 samples in Boston).

Significance

This study joins clinical hematologists, hematopathologists, and computational scientists. The slide digitization and data infrastructure to support this study have been established highlighting the ambition, statistical power, and realism of the proposal. We will translate the results into a medical software to be able to identify patients with a risk of developing MDS or AML at the population level. Beyond this study, therapeutic measures could be developed to even prevent or slow down development of leukemia. In addition, we will discover how cell counts and their shape of MDS patients could inform physicians in treatment decisions. By breaking down the AI models, we will better understand the connection between cell patterns, gene rearrangement and disease outcome. Collectively, results of this novel transatlantic collaboration will be visibly communicated in high-impact scientific publications and activity at international conferences.