Sparse autoencoder for unsupervised nucleus detection and representation in histopathology images

Abstract

Histopathology images are crucial to the study of complex diseases such as cancer. The histologic characteristics of nuclei play a key role in disease diagnosis, prognosis and analysis. In this work, we propose a sparse Convolutional Autoencoder (CAE) for fully unsupervised, simultaneous nucleus detection and feature extraction in histopathology tissue images. Our CAE detects and encodes nuclei in image patches in tissue images into sparse feature maps that encode both the location and appearance of nuclei. Our CAE is the first unsupervised detection network for computer vision applications. The pretrained nucleus detection and feature extraction modules in our CAE can be finetuned for supervised learning in an end-to-end fashion. We evaluate our method on four datasets and reduce the errors of state-of-the-art methods up to 42%. We are able to achieve comparable performance with only 5% of the fullysupervised annotation cost.

Spatial Organization and Molecular Correlation of Tumor-Infiltrating Lymphocytes Using Deep Learning on Pathology Images

“…The necrosis CNN was retrained only once in this study, because it achieved sufficient prediction accuracy. The training and re-training steps of both CNNs involve cross-validation to assess prediction performance and avoid overfitting (Hou et al, 2017). See the Method Details for an in-depth description of this process.…”

“…The CNN and the CAE are designed to have relatively high resolution input such that one can recognize individual lymphocytes. We have chosen to apply unsupervised CAE pre-training because many studies have shown that it boosts the performance of the CNN, please refer to our technical report (Hou et al, 2017). Using the lung adenocarcinoma (LUAD) patches, we empirically showed that the CNN without pre-training achieved significantly lower area under the curve (AUC).…”

“…We then apply global average pooling on the two branches. The pooled features are then concatenated together, followed by a final classification layer with sigmoid activation function (Hou et al, 2017). …”

“…We then trained the CNNs with labeled images. The training phases of the CNNs involve a cross-validation step to assess prediction performance and avoid overfitting (Hou et al, 2017). …”

Spatial heterogeneity and organization of tumor mutation burden and immune infiltrates within tumors based on whole slide images correlated with patient survival in bladder cancer

“…Visually examination of H&E stained histopathology slides by pathologists is a standard clinical practice to diagnose and grade different type of cancers. Many previous studies have been focusing on detection/segmentation of tissue objects [33] [34], or extracting histological image features to assist pathologists in computer-aided tumor diagnosis [5] [11]. Until recently, there are two existing studies [15][17] which explored and found that some gene mutations (e.g., SPOP, KRAS mutations) are predictable by using histological image features in prostate and lung cancer pathology slides.…”