Showing 1–1 of 1 results for author: Pirovano, A
-
Regression Constraint for an Explainable Cervical Cancer Classifier
Authors:
Antoine Pirovano,
Leandro G. Almeida,
Said Ladjal
Abstract:
This article adresses the problem of automatic squamous cells classification for cervical cancer screening using Deep Learning methods. We study different architectures on a public dataset called Herlev dataset, which consists in classifying cells, obtained by cervical pap smear, regarding the severity of the abnormalities they represent. Furthermore, we use an attribution method to understand whi…
▽ More
This article adresses the problem of automatic squamous cells classification for cervical cancer screening using Deep Learning methods. We study different architectures on a public dataset called Herlev dataset, which consists in classifying cells, obtained by cervical pap smear, regarding the severity of the abnormalities they represent. Furthermore, we use an attribution method to understand which cytomorphological features are actually learned as discriminative to classify severity of the abnormalities. Through this paper, we show how we trained a performant classifier: 74.5\% accuracy on severity classification and 94\% accuracy on normal/abnormal classification.
△ Less
Submitted 19 August, 2019; v1 submitted 7 August, 2019;
originally announced August 2019.
