Novel Pipeline for Diagnosing Acute Lymphoblastic Leukemia Sensitive to Related Biomarkers

TL;DR

This paper tackles ALL diagnosis from blood smear images under the practical constraint of small medical datasets, which often leads to non-generalizable shortcuts in end-to-end models. It proposes a hematologist-inspired four-step pipeline—object detection of white blood cells, per-cell feature extraction, LSTM-based patient profiling, and final classification—cast as a multiple-instance learning problem. On ALL IDB 1, it achieves 96.15% accuracy and 94.24% F1, with strong sensitivity to blast-cell biomarkers, and demonstrates acceptable out-of-distribution performance on the Raabin Leukemia dataset. The work demonstrates reliable ALL detection with limited data and outlines how a structured, biomarker-focused approach can generalize to other small medical datasets, while also identifying staining sensitivity as a key area for improvement.

Abstract

Acute Lymphoblastic Leukemia (ALL) is one of the most common types of childhood blood cancer. The quick start of the treatment process is critical to saving the patient's life, and for this reason, early diagnosis of this disease is essential. Examining the blood smear images of these patients is one of the methods used by expert doctors to diagnose this disease. Deep learning-based methods have numerous applications in medical fields, as they have significantly advanced in recent years. ALL diagnosis is not an exception in this field, and several machine learning-based methods for this problem have been proposed. In previous methods, high diagnostic accuracy was reported, but our work showed that this alone is not sufficient, as it can lead to models taking shortcuts and not making meaningful decisions. This issue arises due to the small size of medical training datasets. To address this, we constrained our model to follow a pipeline inspired by experts' work. We also demonstrated that, since a judgement based on only one image is insufficient, redefining the problem as a multiple-instance learning problem is necessary for achieving a practical result. Our model is the first to provide a solution to this problem in a multiple-instance learning setup. We introduced a novel pipeline for diagnosing ALL that approximates the process used by hematologists, is sensitive to disease biomarkers, and achieves an accuracy of 96.15%, an F1-score of 94.24%, a sensitivity of 97.56%, and a specificity of 90.91% on ALL IDB 1. Our method was further evaluated on an out-of-distribution dataset, which posed a challenging test and had acceptable performance. Notably, our model was trained on a relatively small dataset, highlighting the potential for our approach to be applied to other medical datasets with limited data availability.

Figures (7)

• Figure 1: Visualization of model weaknesses. The attention map shows the model focusing on cell boundaries and artifacts, rather than the nucleus, indicating potential limitations in the model's ability to accurately identify relevant features.
• Figure 2: Proposed pipeline for automatic diagnosis of acute lymphoblastic leukemia (ALL) using microscopic images. The pipeline consists of four main steps: (1) Object Detection to detect white blood cells in the input image, (2) Feature Extraction (3) Profiling using an LSTM, and (4) Final Classification. This pipeline enables us to mimic the decision-making process used by hematologists and provides a reliable and accurate way to diagnose ALL using microscopic images.
• Figure 3: Example of an image series from the generated dataset for LSTM network training. This sample is from the cancer class and contains blank, blast cell, and normal cell images with augmentations
• Figure 4: Sample of the modified ALL IDB 1 dataset used for model evaluation. Black squares indicate the removed cells from the ALL IDB 2 dataset used in training to ensure a valid evaluation. Furthermore, we ensured that each image labeled as cancer contained at least one blast cell.
• Figure 5: Effect of removing blast cells and normal cells on the model's recall. The removal of blast cells decreases the model recall by an average of 18.84% across all group sizes, while the removal of normal cells increases the model recall by an average of 18.69%. The results validate our hypothesis and demonstrate the significance of blast cells in the model's decision-making process.