Ordinal losses for classification of cervical cancer risk

Abstract

Cervical cancer is the fourth leading cause of cancer-related deaths in women, especially in low to middle-income countries. Despite the outburst of recent scientific advances, there is no totally effective treatment, especially when diagnosed in an advanced stage. Screening tests, such as cytology or colposcopy, have been responsible for a substantial decrease in cervical cancer deaths. Cervical cancer automatic screening via Pap smear is a highly valuable cell imaging-based detection tool, where cells must be classified as being within one of a multitude of ordinal classes, ranging from abnormal to normal. Current approaches to ordinal inference for neural networks are found to not sufficiently take advantage of the ordinal problem or to be too uncompromising. A non-parametric ordinal loss for neuronal networks is proposed that promotes the output probabilities to follow a unimodal distribution. This is done by imposing a set of different constraints over all pairs of consecutive labels which allows for a more flexible decision boundary relative to approaches from the literature. Our proposed loss is contrasted against other methods from the literature by using a plethora of deep architectures. A first conclusion is the benefit of using non-parametric ordinal losses against parametric losses in cervical cancer risk prediction. Additionally, the proposed loss is found to be the top-performer in several cases. The best performing model scores an accuracy of 75.6% for seven classes and 81.3% for four classes.

Cervical Cancer Classification Using Combined Machine Learning and Deep Learning Approach

“…The Herlev Dataset was used in this study. [4] In article by Albuquerque et al [4], the researcher has proposed a new non-parametric loss for the multi-class Pap smear cell classification. This proposed method was based on a convolutional neural network.…”

“…The study is conducted on the Herlev Pap Smear dataset that is consisted of 917 cell images, and each one contains one single nucleus. The dataset was collected by Herlev University Hospital (Denmark) and the Technical University of Denmark [4]. These cervical cell images are collected manually, which are then annotated into seven classes by skilled cyto-technicians.…”

Uncertainty Ordinal Multi-Instance Learning for Breast Cancer Diagnosis

“…Ordinal classification problems [ 15 , 26 , 27 , 28 ] can be viewed as an intermediate problem between classification and regression, where the target variable is both categorical and ordinal. In general classification problems, the categorical variables are taken from a finite set, and there is no metric relationship between the categories, although they are represented numerically.…”

Unimodal Distributions for Ordinal Regression

“…Among methods recently developed specifically for ordinal regression, unimodal constraints on the probability distribution of the output have proven to be strong contenders. This paper identifies the limitations of current approaches that rely on unimodal distributions [1], [2], [3], [4], and addresses these issues through two novel solutions, one that imposes hard constraints on the distribution with a new neural network architecture and the other that promotes unimodal distributions through a loss term that penalizes deviations from unimodality.…”