2024 · May

AI-Driven Fibrosis Analysis Demonstrated in Phase II Clinical Trial

Our Continuous Indexing of Fibrosis (CIF) technology has been demonstrated in a multicenter Phase II clinical trial—the first application of AI-driven quantitative fibrosis analysis in this setting. Published in HemaSphere, the study shows how objective, automated measurement of bone marrow fibrosis can augment traditional manual grading in patients with myelofibrosis.

The challenge with manual fibrosis grading

Bone marrow fibrosis is central to diagnosing and monitoring myeloproliferative neoplasms (MPNs), including myelofibrosis. Clinicians rely on the European Consensus grading system, which assigns samples to categories from MF-0 (no fibrosis) to MF-3 (severe fibrosis). But this approach is subjective and semiquantitative—it cannot fully capture the heterogeneity within a single sample.

Our analysis revealed just how much overlap exists. Among the trial samples, 38% of those graded MF-2 fell within the score distribution of MF-3, and 48% of MF-3 samples overlapped with MF-2. This isn't surprising given the recognized difficulty of distinguishing between these grades, but it raises important questions about relying solely on manual assessment when evaluating antifibrotic therapies.

What CIF adds

CIF uses a convolutional neural network trained on reticulin-stained bone marrow trephine samples. Rather than assigning a single grade, it generates a continuous score from 0 to 1 for each region of tissue, producing a heatmap that captures both overall severity and spatial heterogeneity.

This granularity matters. Some samples with low average fibrosis were correctly graded MF-2 by pathologists because more than 30% of the tissue contained severe fibrosis—something a single average score would miss. CIF captures this heterogeneity, enabling more nuanced comparison across patients and over time.

Results from the zinpentraxin alfa trial

We analyzed 142 bone marrow samples from 50 patients enrolled in a Phase II study of zinpentraxin alfa, an antifibrotic therapy. Samples were collected at screening, cycle 4, and cycle 9.

Key findings:

• 38% of patients showed improvement in average CIF score by cycle 9
• Patients with higher baseline CIF scores were significantly more likely to show fibrosis reduction (p < 0.01)
• There was a trend toward association between CIF improvement and clinical response per IWG-MRT criteria
• Only 15% of cases showed concordance between manual grade improvement and CIF score improvement—highlighting the limitations of subjective grading

Implications for clinical trials

These results suggest that variation in manual fibrosis assessment could affect the accuracy and consistency of trials evaluating therapies targeting myelofibrosis. Quantitative analysis like CIF provides an objective measure that can be compared across sites, timepoints, and patient cohorts.

This work was a collaboration with F. Hoffmann-La Roche, Genentech, and expert hematopathologists from Massachusetts General Hospital, University of Pennsylvania, and UT Southwestern, with clinical input from Guy's & St Thomas' NHS Foundation Trust.

Read the full paper in HemaSphere: DOI: 10.1002/hem3.105