# On evaluating CNN representations for low resource medical image classification

## TL;DR

This paper explores the use of CNN transfer learning for low-resource medical image classification, introduces a metric to predict CNN performance, and demonstrates its effectiveness in gastrointestinal landmark classification with limited data.

## Contribution

It presents a novel metric to predict CNN test performance from training set representations and evaluates CNN transfer learning in low-resource medical imaging tasks.

## Key findings

- CNN transfer learning outperforms knowledge-driven features.
- The proposed metric correlates 87% with actual test performance.
- CNN choice impacts classification success in low-resource settings.

## Abstract

Convolutional Neural Networks (CNNs) have revolutionized performances in several machine learning tasks such as image classification, object tracking, and keyword spotting. However, given that they contain a large number of parameters, their direct applicability into low resource tasks is not straightforward. In this work, we experiment with an application of CNN models to gastrointestinal landmark classification with only a few thousands of training samples through transfer learning. As in a standard transfer learning approach, we train CNNs on a large external corpus, followed by representation extraction for the medical images. Finally, a classifier is trained on these CNN representations. However, given that several variants of CNNs exist, the choice of CNN is not obvious. To address this, we develop a novel metric that can be used to predict test performances, given CNN representations on the training set. Not only we demonstrate the superiority of the CNN based transfer learning approach against an assembly of knowledge driven features, but the proposed metric also carries an 87% correlation with the test set performances as obtained using various CNN representations.