Impact of JPEG 2000 compression on deep convolutional neural networks for metastatic cancer detection in histopathological images

Abstract

The availability of massive amounts of data in histopathological whole-slide images (WSIs) has enabled the application of deep learning models and especially convolutional neural networks (CNNs), which have shown a high potential for improvement in cancer diagnosis. However, storage and transmission of large amounts of data such as gigapixel histopathological WSIs are challenging. Exploiting lossy compression algorithms for medical images is controversial but, as long as the clinical diagnosis is not affected, is acceptable. We study the impact of JPEG 2000 compression on our proposed CNN-based algorithm, which has produced performance comparable to that of pathologists and which was ranked second place in the CAMELYON17 challenge. Detecting tumor metastases in hematoxylin and eosin-stained tissue sections of breast lymph nodes is evaluated and compared with the pathologists' diagnoses in three different experimental setups. Our experiments show that the CNN model is robust against compression ratios up to 24:1 when it is trained on uncompressed high-quality images. We demonstrate that a model trained on lower quality images-i.e., lossy compressed images-depicts a classification performance that is significantly improved for the corresponding compression ratio. Moreover, it is also observed that the model performs equally well on all higher-quality images. These properties will help to design cloud-based computer-aided diagnosis (CAD) systems, e.g., telemedicine that employ deep CNN models that are more robust to image quality variations due to compression required to address data storage and transmission constraints. However, the results presented are specific to the CAD system and application described, and further work is needed to examine whether they generalize to other systems and applications.

Quantitative Assessment of the Effects of Compression on Deep Learning in Digital Pathology Image Analysis

“…Our approach in this study differed from that of Zanjani et al 11 and Doyle et al 29 in the following 2 ways. Firstly, our study and that by Zanjani et al 11 assessed the impact of lossy compression on DL, whereas Doyle et al 29 explored the impact of JPEG2000 compression on a handcrafted machine learning approach. Compared with the work of Doyle et al 29 , the DL models were shown to be less robust to compression artifacts than their CAD system.…”

“… 5 Although recent research has investigated the effects of compression on DL performance in natural images, 15 relatively little study has taken place in dp-based image analysis tasks. 11 , 29 We aimed to address that need by studying the inverse relationship between compression and performance of DL algorithms in DP images.…”

“…A recent study by Zanjani et al 11 evaluated the impact of JPEG2000 compression on a DL model for slide-based breast lymph node metastasis detection. Our study confirms their findings: their DL models seemed to be robust up to a compression ratio of 1:24 (96% reduction in size).…”

“…Although there has recently been a great deal of interest in developing and applying DL approaches in DP, the question of the effect of image compression on DL algorithms has been largely unexplored. 11 Compression technologies are especially important in countries with poor quality Internet access, where sending and receiving large DP image files can be challenging. 12 , 13 For instance, a single prostate biopsy slide digitized at 40× can easily result in > 5 gigabytes of data, with a typical pathology workflow requiring approximately 12 slides.…”

Recognition of Cutaneous Melanoma on Digitized Histopathological Slides via Artificial Intelligence Algorithm

“…In recent years, few studies that focus on deep learning algorithms have been proposed to automate the analysis of melanoma and skin lesions in WSIs ( 17 – 19 ). Since the sizes of WSIs are too large to be used as direct input to a CNN, the typical approach is to train, validate, and test the CNN, instead of using low-pixel-resolution patches of the WSI, obtaining tens to thousands of patches from each WSI ( 20 ). Although AI technology has achieved remarkable results for skin pathology analysis, in this field, the potential of CNNs has not been fully investigated, and their performances may be significantly improved.…”