Article
Fast Pre-Diagnosis of Neoplastic Changes in Cytology Images
Using Machine Learning
Jakub Caputa
1
, Daria Łukasik
1
, Maciej Wielgosz
1,2,
* , Michał Karwatowski
1,2
, Rafał Fr ˛aczek
1,2
,
Paweł Russek
1,2
and Kazimierz Wiatr
1,2
Citation: Caputa, J.; Łukasik, D.;
Wielgosz, M.; Karwatowski, M.;
Fr ˛aczek, R.; Russek, P.; Wiatr, K. Fast
Pre-Diagnosis of Neoplastic Changes
in Cytology Images Using Machine
Learning. Appl. Sci. 2021, 11, 7181.
https://doi.org/10.3390/app11167181
Academic Editor: Keun Ho Ryu
Received: 10 July 2021
Accepted: 30 July 2021
Published: 4 August 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
1
Academic Computer Centre CYFRONET AGH, ul. Nawojki 11, 30-072 Kraków, Poland;
jjakubcaputa@gmail.com (J.C.); daria.lukasik.vet@gmail.com (D.Ł.); mkarwat@agh.edu.pl (M.K.);
rafalfr@agh.edu.pl (R.F.); russek@agh.edu.pl (P.R.); wiatr@agh.edu.pl (K.W.)
2
Institute of Electronics, AGH University of Science and Technology, al. Adama Mickiewicza 30,
30-059 Kraków, Poland
* Correspondence: wielgosz@agh.edu.pl
Abstract:
We present the experiment results to use the YOLOv3 neural network architecture to
automatically detect tumor cells in cytological samples taken from the skin in canines. A rich dataset
of 1219 smeared sample images with 28,149 objects was gathered and annotated by the vet doctor
to perform the experiments. It covers three types of common round cell neoplasms: mastocytoma,
histiocytoma, and lymphoma. The dataset has been thoroughly described in the paper and is publicly
available. The YOLOv3 neural network architecture was trained using various schemes involving
original dataset modification and the different model parameters. The experiments showed that
the prototype model achieved 0.7416 mAP, which outperforms the state-of-the-art machine learning
and human estimated results. We also provided a series of analyses that may facilitate ML-based
solutions by casting more light on some aspects of its performance. We also presented the main
discrepancies between ML-based and human-based diagnoses. This outline may help depict the
scenarios and how the automated tools may support the diagnosis process.
Keywords: canines; neoplasms; detection; deep learning; YOLOv3
1. Introduction
Veterinary oncology is a medical science field in which a precise diagnosis of the
examined physical condition before introducing treatment is crucial for its effects and
allows a doctor for a reasonable decision to be taken regarding the treatment of an oncolog-
ical patient.
Many diagnostic methods (including clinical examination, imaging tests, or endo-
scopic examination) allow examiners for excellent visualization of the lesions, as well as to
recognize their structure, size, number, and features of clinical malignancy (rapid growth,
large volume of lesion, binding to oral tissues), invasive, infiltrative nature of growth, and
destruction of adjacent structures [
1
]. A microscopic examination of tissue samples allows
us to determine actual tumor nature [1].
Unfortunately, in many cases, the process involves having samples shipped to a
dedicated laboratory for the examination, requiring additional time that may contribute
to the deterioration of the patient’s state. Alternatively, the microscopic examination may
be conducted by a site physician. However, this approach is prone to errors, and not all
doctors have appropriate training. Automating the initial diagnosis process using artificial
intelligence methods (AI) may help make the process smoother.
Consequently, this paper aims to present a performance of the proof of concept system
designed for automatic detection of neoplastic cells in cytology samples from canine skin
tumors. The system is meant to ease the work of veterinarians and significantly shorten
the time of diagnosis. The goal was also to prepare the auxiliary software tools that would
Appl. Sci. 2021, 11, 7181. https://doi.org/10.3390/app11167181 https://www.mdpi.com/journal/applsci
