Deep convolutional neural network applied to Trypanosoma cruzi detection in blood samples
by André S. Pereira; Leonardo O. Mazza; Pedro C.C. Pinto; José Gabriel R.C. Gomes; Nadia Nedjah; Daniel F. Vanzan; Alexandre S. Pyrrho; Juliana G.M. Soares
International Journal of Bio-Inspired Computation (IJBIC), Vol. 19, No. 1, 2022

Abstract: Standard diagnosis of Chagas disease, during its acute phase, is based on Trypanosoma cruzi visualisation through microscopy applied to peripheral blood slides. We apply MobileNet V2 convolutional layers to image tiles from acute-phase peripheral blood samples, and feed the resulting 1,280-dimensional feature vectors into a single-neuron classifier. Using sample image tiles from a reduced, 12-slide, dataset, we achieve accuracy equal to 96.4% on a balanced validation subset. On image tiles from a 13th blood smear slide, test accuracy is estimated at 72.0%. We extend the dataset with images from six additional slides, which include two thick blood samples, and then test accuracy on two extra slides improves to 95.4%. Examples of raster scans with overlapping windows illustrate the detection of all positive instances of Trypanosoma cruzi in blood smear and thick blood images, without any false alarm. Furthermore, we start an investigation of boosting algorithm effects on classifier performance.

Online publication date: Mon, 07-Feb-2022

The full text of this article is only available to individual subscribers or to users at subscribing institutions.

 
Existing subscribers:
Go to Inderscience Online Journals to access the Full Text of this article.

Pay per view:
If you are not a subscriber and you just want to read the full contents of this article, buy online access here.

Complimentary Subscribers, Editors or Members of the Editorial Board of the International Journal of Bio-Inspired Computation (IJBIC):
Login with your Inderscience username and password:

Username:        Password:          Forgotten your password?

Want to subscribe?
A subscription gives you complete access to all articles in the current issue, as well as to all articles in the previous three years (where applicable). See our Orders page to subscribe.

If you still need assistance, please email subs@inderscience.com