Structure Prediction for Gland Segmentation with Hand-Crafted and Deep Convolutional Features

TY - JOUR

T1 - Structure Prediction for Gland Segmentation with Hand-Crafted and Deep Convolutional Features

AU - Manivannan, Siyamalan

AU - Li, Wenqi

AU - Zhang, Jianguo

AU - Trucco, Emanuele

AU - McKenna, Stephen

N1 - no funding info

PY - 2018/1

Y1 - 2018/1

N2 - We present a novel method to segment instances of glandular structures from colon histopathology images. We use a structure learning approach which represents local spatial configurations of class labels, capturing structural information normally ignored by sliding-window methods. This allows us to reveal different spatial structures of pixel labels (e.g., locations between adjacent glands, or far from glands), and to identify correctly neighbouring glandular structures as separate instances. Exemplars of label structures are obtained via clustering and used to train support vector machine classifiers. The label structures predicted are then combined and post-processed to obtain segmentation maps. We combine hand-crafted, multi-scale image features with features computed by a deep convolutional network trained to map images to segmentation maps. We evaluate the proposed method on the public domain GlaS dataset, which allows extensive comparisons with recent, alternative methods. Using the GlaS contest protocol, our method achieves the overall best performance.

AB - We present a novel method to segment instances of glandular structures from colon histopathology images. We use a structure learning approach which represents local spatial configurations of class labels, capturing structural information normally ignored by sliding-window methods. This allows us to reveal different spatial structures of pixel labels (e.g., locations between adjacent glands, or far from glands), and to identify correctly neighbouring glandular structures as separate instances. Exemplars of label structures are obtained via clustering and used to train support vector machine classifiers. The label structures predicted are then combined and post-processed to obtain segmentation maps. We combine hand-crafted, multi-scale image features with features computed by a deep convolutional network trained to map images to segmentation maps. We evaluate the proposed method on the public domain GlaS dataset, which allows extensive comparisons with recent, alternative methods. Using the GlaS contest protocol, our method achieves the overall best performance.

KW - Molecular and cellular imaging

KW - Gastrointestinal tract

KW - Segmentation

KW - gastrointestinal tract

KW - segmentation

KW - Histocytochemistry/methods

KW - Colorectal Neoplasms/diagnostic imaging

KW - Colon/diagnostic imaging

KW - Humans

KW - Intestinal Mucosa/diagnostic imaging

KW - Adenocarcinoma/diagnostic imaging

KW - Support Vector Machine

KW - Molecular Imaging/methods

KW - Image Processing, Computer-Assisted/methods

UR - https://www.scopus.com/pages/publications/85040683314

U2 - 10.1109/TMI.2017.2750210

DO - 10.1109/TMI.2017.2750210

M3 - Article

C2 - 28910760

SN - 0278-0062

VL - 37

SP - 210

EP - 221

JO - IEEE Transactions on Medical Imaging

JF - IEEE Transactions on Medical Imaging

IS - 1

ER -