Author Date

2026-08-01

Degree Name

BS

Department

Mathematics

College

Computational, Mathematical and Physical Sciences

Defense Date

2026-05-11

Publication Date

2026-08-07

First Faculty Advisor

Emily Evans

Second Faculty Advisor

Ryan Cordner

First Faculty Reader

Tyler Jarvis

Honors Coordinator

Davi Obata

Keywords

Hemophilia, Factor Deficiency, Screening, Coagulation, Image Classification, Neural Network

Abstract

When blood vessels are damaged, clots formed by the protein fibrin stabilize platelets and stem active bleeding. Coagulation, the physiological process that forms fibrin clots from blood plasma, hinges on the clotting cascade, a complex network of interacting proteins. These so-called clotting factors help control fibrin polymerization and thereby coagulation. A clotting factor deficiency generally slows coagulation, resulting in bleeding disorders like hemophilia A, B or C. Diagnosing the deficient factor's identity is essential to optimal treatment. Currently, missing factors are determined in reference laboratories via specialized tests. A more accessible test could help streamline the diagnosis process--in this work, we present a possible candidate. We distinguish between missing clotting factors via brightfield microscopy inspection of partial thromboplastin time (PTT)-induced clots. Our public dataset contains 7,400 brightfield microscopy images of clotted pooled plasmas of five phenotypes: normal and prolonged clotting and deficiencies in clotting factors VIII, IX, and XI. Images were sampled from slide surfaces using two distinct methods (n=1000 and n=6400). We form a holdout set, develop a convolutional neural network ensemble using the remaining data, and classify the holdout images from the first sampling method (n=200) with 82.5% overall subset accuracy and with at least 65% precision on all five classes. Despite consensus in broader literature that the identity of the deficient clotting factor affects clot structural properties, we found no other scientific works utilizing microscopic images of fibrin clots to identify missing factors. Our dataset is unique, and our results provide a benchmark for future work.

Download

Share

COinS