Detecting Cancer Metastases on Gigapixel Pathology Images

Abstract

Each year, the treatment decisions for more than 230, 000 breast cancer patients in the U.S. hinge on whether the cancer has metastasized away from the breast. Metastasis detection is currently performed by pathologists reviewing large expanses of biological tissues. This process is labor intensive and error-prone. We present a framework to automatically detect and localize tumors as small as 100 × 100 pixels in gigapixel microscopy images sized 100, 000×100, 000 pixels. Our method leverages a convolutional neural network (CNN) architecture and obtains state-of-the-art results on the Camelyon16 dataset in the challenging lesion-level tumor detection task. At 8 false positives per image, we detect 92.4% of the tumors, relative to 82.7% by the previous best automated approach. For comparison, a human pathologist attempting exhaustive search achieved 73.2% sensitivity. We achieve image-level AUC scores above 97% on both the Camelyon16 test set and an independent set of 110 slides. In addition, we discover that two slides in the Came-lyon16 training set were erroneously labeled normal. Our approach could considerably reduce false negative rates in metastasis detection.

Supervised Transfer Learning at Scale for Medical Imaging

“…For example, the chest X-Ray model follows a standard paradigm of predicting from features extracted from a single image, but the mammography model concatenates features extracted from four different images, and the dermatology model averages features from a variable number (1-6) of images. We note that these models have significantly distinct setups, however, these setups are necessary to be able to draw meaningful conclusions at state-of-the-art quality levels [34,30]. Full details of the task-specific setups are in Appendix B.1.…”

“…Deep learning has enabled many exciting recent advances to medical imaging. High performing models have been developed, often competitive with human experts, in a variety of domains including ophthalmology [14], radiology [34,39], dermatology [10,31] and pathology [30].…”

What Can Machine Vision Do for Lymphatic Histopathology Image Analysis: A Comprehensive Review

“…ResNeXt is a deep CNN with 101 layers and has a characteristic of stability and high efficiency. Next, an ensemble learning approach is applied for two classification tasks including 5 In [189], a framework is proposed to automatically detect and localize tumors of small pixels with 100×100 in microscopy images with a size of 100,000×100,000. A CNN architecture called InceptionV3 is utilized to carry out the detection task in the Camelyon16 dataset.…”

“…the residual CNN is used to classify lymphomas subtypes, an ACC of 97.67% which is higher than U-net and ResNet is obtained [167]. 92.4% of tumors are detected through a CNN architecture called Inception V3 in [189]. CNN obtains good results on histopathological images of lymphoma, but also on other pathological images, such as breast cancer [194,195], prostate cancer [196,197], colon cancer [198,199].…”

An Efficient Cervical Whole Slide Image Analysis Framework Based on Multi-scale Semantic and Spatial Deep Features

“…The public CAMELYON datasets ( [19]) consists of 1,399 annotated WSIs of lymph nodes to detect breast cancer also promoted the development. The grand challenge CAMELYON17 has 37 algorithms to predict at the WSI level using 899 learnable WSIs, where the top performance from the GoogleNet was reproduced based on the multi-scale and color normalization ( [20]), which with rela-tively shadow architecture and relatively longer training phase than others. After the patch level computation with CNN, this top method used the conventional machine learning and feature engineering to classify WSIs, which was surpassed by PFA-ScanNet that using more scales to extract features ( [21]), then it was surpassed by the novel attention-based classifier with a shadower siamese MI-FCN architecture ( [22]).…”