Neural Image Compression for Gigapixel Histopathology Image Analysis
This addresses the challenge of reducing annotation burden for gigapixel image analysis in medical domains, though it is incremental as it builds on existing compression and CNN techniques.
The paper tackled the problem of analyzing gigapixel histopathology images without fine-grained annotations by proposing Neural Image Compression (NIC), a two-step method that compresses images using unsupervised neural networks and trains a CNN on compressed representations to predict image-level labels, achieving successful integration of global and local visual cues and confirmation of attention regions overlapping with expert annotations.
We propose Neural Image Compression (NIC), a two-step method to build convolutional neural networks for gigapixel image analysis solely using weak image-level labels. First, gigapixel images are compressed using a neural network trained in an unsupervised fashion, retaining high-level information while suppressing pixel-level noise. Second, a convolutional neural network (CNN) is trained on these compressed image representations to predict image-level labels, avoiding the need for fine-grained manual annotations. We compared several encoding strategies, namely reconstruction error minimization, contrastive training and adversarial feature learning, and evaluated NIC on a synthetic task and two public histopathology datasets. We found that NIC can exploit visual cues associated with image-level labels successfully, integrating both global and local visual information. Furthermore, we visualized the regions of the input gigapixel images where the CNN attended to, and confirmed that they overlapped with annotations from human experts.