[Paper Review] Using Convolutional Neural Networks for Determining Reticulocyte Percentage in Cats

Krunoslav Vinicki, Pierluigi Ferrari+2 more|arXiv (Cornell University)|Mar 13, 2018

Cell Image Analysis Techniques6 references19 citations

TL;DR

This paper proposes a deep learning approach using a pre-trained Single-Shot MultiBox Detector (SSD) in Keras to automatically count reticulocytes in feline blood smear images. Trained on only 800 labeled images, the model achieves 98.7% accuracy in predicting reticulocyte counts, demonstrating that accessible, high-accuracy AI solutions are feasible even in veterinary medicine with limited data.

ABSTRACT

Recent advances in artificial intelligence (AI), specifically in computer vision (CV) and deep learning (DL), have created opportunities for novel systems in many fields. In the last few years, deep learning applications have demonstrated impressive results not only in fields such as autonomous driving and robotics, but also in the field of medicine, where they have, in some cases, even exceeded human-level performance. However, despite the huge potential, adoption of deep learning-based methods is still slow in many areas, especially in veterinary medicine, where we haven't been able to find any research papers using modern convolutional neural networks (CNNs) in medical image processing. We believe that using deep learning-based medical imaging can enable more accurate, faster and less expensive diagnoses in veterinary medicine. In order to do so, however, these methods have to be accessible to everyone in this field, not just to computer scientists. To show the potential of this technology, we present results on a real-world task in veterinary medicine that is usually done manually: feline reticulocyte percentage. Using an open source Keras implementation of the Single-Shot MultiBox Detector (SSD) model architecture and training it on only 800 labeled images, we achieve an accuracy of 98.7% at predicting the correct number of aggregate reticulocytes in microscope images of cat blood smears. The main motivation behind this paper is to show not only that deep learning can approach or even exceed human-level performance on a task like this, but also that anyone in the field can implement it, even without a background in computer science.

Motivation & Objective

• To develop an accessible, deep learning-based method for automated reticulocyte counting in cat blood smears.
• To demonstrate that high-accuracy results can be achieved with minimal labeled data in veterinary medical imaging.
• To lower the barrier to entry for non-computer scientists in veterinary medicine by using open-source tools like Keras.
• To validate that deep learning can match or exceed human-level performance in a real-world veterinary diagnostic task.
• To promote wider adoption of AI in veterinary medicine by proving feasibility with modest computational and data requirements.

Proposed method

• The authors used an open-source Keras implementation of the Single-Shot MultiBox Detector (SSD) architecture for object detection.
• The model was fine-tuned on a dataset of 800 labeled microscope images of feline blood smears containing reticulocytes.
• Reticulocytes were treated as objects to be detected and counted, with bounding boxes used to localize each cell.
• The training process employed transfer learning and data augmentation to improve generalization on a small dataset.
• Model evaluation was performed using standard classification accuracy metrics on a held-out test set.
• The implementation was designed to be reproducible and accessible to users without advanced computer science expertise.

Experimental results

Research questions

• RQ1 Can a deep learning model achieve high accuracy in counting reticulocytes in feline blood smears using only 800 labeled images?
• RQ2 Can a pre-trained SSD model be effectively fine-tuned for a veterinary medical imaging task with limited data?
• RQ3 Is it feasible for non-computer scientists in veterinary medicine to implement and use such a model with open-source tools?
• RQ4 Does the model's performance exceed or match human-level accuracy in reticulocyte counting?
• RQ5 Can this approach serve as a scalable, low-cost alternative to manual reticulocyte counting in clinical settings?

Key findings

• The model achieved a test accuracy of 98.7% in predicting the correct number of aggregate reticulocytes in blood smear images.
• The system demonstrated high reliability despite being trained on only 800 labeled images, indicating strong generalization with limited data.
• The use of transfer learning and the Keras-based SSD framework enabled effective training without requiring extensive computational resources.
• The results suggest that deep learning models can achieve human-level or better performance in veterinary reticulocyte counting tasks.
• The implementation is accessible to non-experts, as the authors emphasize usability through open-source tools and minimal technical requirements.
• This study presents a viable pathway for integrating AI into veterinary diagnostics with minimal data and technical barriers.