•  
  •  
 
SACAD: Scholarly Activities

Abstract

Deep learning shows strong potential in medical-image analysis, yet adoption in cyptopathology

remains limited. Cytopathology could benefit from deep learning applications by improving

diagnostic efficiency and accuracy. However deep learning comes with a notorious “black box”

that keeps the models from being transparent and trustworthy for widespread clinical adoption.

We conducted a comprehensive and comparative analysis of several deep learning architectures

for multi-class classification of acute leukemia types, ALL, AML, and normal healthy cells from

peripheral blood smear images. The models in this research include a Vision Transformer (ViT)

and a diverse selection of Convolutional Neural Network (CNN) models. The performance of the

models was assessed using F1-score, AUC, precision, and recall. The top model achieving an

F1-score of 99.94%. In order to address the “black box” issue a multi-faceted explainable AI

(XAI) analysis was performed by utilizing methods such as Grad-CAM, Score-CAM, LIME, and

Integrated Gradients. These techniques provide visual explanations that present a qualitative

comparison of the architectures’ feature attribution. This showcases the cellular morphologies

that were weighted the most for any given prediction. This research seeks to provide an

evaluation of model performance as well as a crucial analysis of interpretability layers that can

contribute to increased trustworthiness and transparency for the use of deep learning in

diagnostics.

Relevance to Kansas: Kansas has a higher rate of leukemia cases than the national average.

Acute leukemia affects both children and adults and is the most aggressive form of leukemia.

Faculty Advisor

Dr. Anas Hourani

Department/Program

Computer Science

Submission Type

online only poster

Date

4-12-2026

Rights

Copyright the Author(s)

Download

Share

COinS