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

Abstract

PURPOSE Deep learning (DL), a class of approaches involving self-learned discriminative features, is increasingly being applied to digital pathology (DP) images for tasks such as disease identification and segmentation of tissue primitives (eg, nuclei, glands, lymphocytes). One application of DP is in telepathology, which involves digitally transmitting DP slides over the Internet for secondary diagnosis by an expert at a remote location. Unfortunately, the places benefiting most from telepathology often have poor Internet quality, resulting in prohibitive transmission times of DP images. Image compression may help, but the degree to which image compression affects performance of DL algorithms has been largely unexplored. METHODS We investigated the effects of image compression on the performance of DL strategies in the context of 3 representative use cases involving segmentation of nuclei (n = 137), segmentation of lymph node metastasis (n = 380), and lymphocyte detection (n = 100). For each use case, test images at various levels of compression (JPEG compression quality score ranging from 1-100 and JPEG2000 compression peak signal-to-noise ratio ranging from 18-100 dB) were evaluated by a DL classifier. Performance metrics including F1 score and area under the receiver operating characteristic curve were computed at the various compression levels. RESULTS Our results suggest that DP images can be compressed by 85% while still maintaining the performance of the DL algorithms at 95% of what is achievable without any compression. Interestingly, the maximum compression level sustainable by DL algorithms is similar to where pathologists also reported difficulties in providing accurate interpretations. CONCLUSION Our findings seem to suggest that in low-resource settings, DP images can be significantly compressed before transmission for DL-based telepathology applications.

Quality control stress test for deep learning-based diagnostic model in digital pathology

“…Interestingly, the human eye does not recognize image alteration within a wide range of compression levels (30-90%). The results of the current study and other published studies [17,33] show that this might have potential consequences for accuracy of analytical models trained on datasets with lower compression levels, should they be applied to such images. Our findings show that any compression levels under 80% can result in accuracy deterioration and should be avoided.…”

“…Several published studies systematically addressed the influence of histological artifacts on performance of DL-based models in diagnostic pathology. A solid basis of evidence is related to focus problem in digital pathology [17,[25][26][27][28]. Most of the studies, however, concentrate on quantitative detection of out-offocus regions using simple DL-based models trained on synthetic data [17,[25][26][27][28] or other principles [29,30].…”

“…The aim of the pathologist and of any assisting system (e.g., DL-based) is to provide top diagnostic accuracy irrespective of pre-analytical issues and artifacts. Only few studies investigate how DL-based models will operate in case of artifacts [17][18][19]. Generalization of this information is difficult for models from different domains, tasks (classification, segmentation, detection), different neural network architectures, and models using different ground truth.…”

Optimization of Quantitative Financial Data Analysis System Based on Deep Learning

“…When the number of trainings is 100, the prediction result is closer to the target result, and the prediction effect is significantly improved compared with one training. When the number of trainings is 1000, the prediction effect is not 6 Complexity significantly improved compared with 100 trainings, indicating that the prediction effect has been saturated and there is no need to increase the amount of training. is paper selects the opening price, closing price, the highest price, the lowest price, and trading volume series of 10 stocks and indexes from 2012 to 2019 as the data set of the model.…”

“…In different academic fields, scholars have proposed a variety of financial analysis and forecasting methods, including (1) financial forecasting based on economics and its improved forecasting methods, such as the financial forecasting method based on grey linear regression combination model, which has the characteristics of less data demand and accurate forecasting model; (2) forecasting methods and models based on mathematics and statistics, such as improved hidden Markov model and its application in financial forecasting, financial forecasting based on Bayesian maximum likelihood estimation, etc. [6,7]. In recent years, there have been some methods to analyze and forecast financial data combined with computer technology.…”