Comparing two classes of end-to-end machine-learning models in lung nodule detection and classification: MTANNs vs. CNNs

作者：

Highlights：

• MTANNs yielded higher performance than CNNs for nodule detection and classification.

• Deep CNN architectures achieved higher performance than shallow architectures for nodule detection.

• CNN architectures with varying depths performed comparably for nodule classification.

• MTANNs can achieve desired performance with a smaller training dataset than do the CNNs.

• MTANNs tend to learn the appearance of lesion parts, whereas CNNs attempt to learn the lesion appearance as a whole.

摘要

Highlights•MTANNs yielded higher performance than CNNs for nodule detection and classification.•Deep CNN architectures achieved higher performance than shallow architectures for nodule detection.•CNN architectures with varying depths performed comparably for nodule classification.•MTANNs can achieve desired performance with a smaller training dataset than do the CNNs.•MTANNs tend to learn the appearance of lesion parts, whereas CNNs attempt to learn the lesion appearance as a whole.

论文关键词：Deep learning,Patch-based machine learning,Image-based machine learning,Massive-training artificial neural network,Convolution neural network,Focal lesions,Classification,Computer-aided diagnosis,Lung nodules

论文评审过程：Received 30 January 2016, Revised 4 September 2016, Accepted 21 September 2016, Available online 7 October 2016, Version of Record 27 November 2016.

论文官网地址：https://doi.org/10.1016/j.patcog.2016.09.029