Deep Learning-Based Sparse Whole-Slide Image Analysis for the Diagnosis of Gastric Intestinal Metaplasia (2201.01449v1)

Abstract: In recent years, deep learning has successfully been applied to automate a wide variety of tasks in diagnostic histopathology. However, fast and reliable localization of small-scale regions-of-interest (ROI) has remained a key challenge, as discriminative morphologic features often occupy only a small fraction of a gigapixel-scale whole-slide image (WSI). In this paper, we propose a sparse WSI analysis method for the rapid identification of high-power ROI for WSI-level classification. We develop an evaluation framework inspired by the early classification literature, in order to quantify the tradeoff between diagnostic performance and inference time for sparse analytic approaches. We test our method on a common but time-consuming task in pathology - that of diagnosing gastric intestinal metaplasia (GIM) on hematoxylin and eosin (H&E)-stained slides from endoscopic biopsy specimens. GIM is a well-known precursor lesion along the pathway to development of gastric cancer. We performed a thorough evaluation of the performance and inference time of our approach on a test set of GIM-positive and GIM-negative WSI, finding that our method successfully detects GIM in all positive WSI, with a WSI-level classification area under the receiver operating characteristic curve (AUC) of 0.98 and an average precision (AP) of 0.95. Furthermore, we show that our method can attain these metrics in under one minute on a standard CPU. Our results are applicable toward the goal of developing neural networks that can easily be deployed in clinical settings to support pathologists in quickly localizing and diagnosing small-scale morphologic features in WSI.

• The paper presents a sparse deep learning framework that rapidly localizes critical regions in whole-slide images for accurate gastric intestinal metaplasia diagnosis.
• It leverages convolutional neural networks to aggregate patch-level classifications, achieving an AUC of 0.98 and an AP up to 0.99 on consumer-grade hardware.
• The approach balances diagnostic performance with computational efficiency, offering a scalable solution for real-time histopathology workflows.

Deep Learning-Based Sparse Whole-Slide Image Analysis for Gastric Intestinal Metaplasia Diagnosis

This paper presents a pioneering approach leveraging deep learning to address the computational challenges inherent in whole-slide image (WSI) analysis for diagnosing gastric intestinal metaplasia (GIM). The authors propose a sparse WSI analytic technique aimed at enhancing the efficiency of detecting regions of interest (ROI) in histopathology by segmenting gigapixel-scale images while maintaining high diagnostic accuracy and reducing inference time.

Overview of Methodology

The authors introduce a method focusing on sparse analysis, whereby only segments of the WSI are analyzed to identify high-power ROI, instead of processing the entire slide. The proposed framework draws inspiration from early classification literature, applying this concept to histopathology. The main objectives are twofold: rapid localization of discriminative morphologic features and robust classification of WSIs for diagnosing GIM.

The sparse approach utilizes convolutional neural networks (CNNs) to sequentially process fixed-sized image patches. These patches, sampled from the slide, undergo classification followed by aggregation of patch-level scores, creating a dynamic update in the slide-level diagnosis. A significant highlight is the adaptability to consumer-grade hardware, which negates the necessity for highly specialized processing equipment and supports deployment in typical clinical environments.

Evaluation Framework

The framework developed quantifies the tradeoff between diagnostic performance and the number of patches processed during inference, enabling effective model selection from candidate models operating at varying magnifications. This evaluation is critical, as it dictates the balance of processing time against the quality of diagnostic outputs, addressing a common bottleneck in clinical histopathology workflows.

Experimental Results

Testing involved a dataset comprising 106 WSIs, split into positive and negative GIM diagnoses. The method achieved an impressive area under the receiver operating characteristic curve (AUC) of 0.98 and an average precision (AP) of 0.95. Notably, the model's diagnostic precision reached an AP of 0.99 within less than one minute using a CPU, marking a significant advancement in reducing clinical inference times for effective deployment. The sparse approach also demonstrated scalability across various levels of training data, maintaining performance robustness without extensive pixel-level annotations.

Implications and Future Prospects

The implications of this research extend both theoretically and practically; it underscores the potential for integrating sparse analysis techniques within broader digital pathology applications. This approach not only optimizes computational resources but also presents an adaptable solution for rapid diagnostic procedures in healthcare environments.

The paper opens avenues for incorporating such methodologies in other diagnostic tasks beyond GIM, where identifying small-scale features efficiently is crucial. It could facilitate the implementation of real-time diagnostic support tools using consumer-grade hardware, further democratizing access to advanced AI in pathology.

As an extension, future work could integrate multi-scale analysis, refining the balance of patch size and resolution, or the exploration of additional preprocessing techniques such as stain normalization. The ongoing development of more sophisticated neural architectures and advances in compute efficiency offers further prospects for enhancing these models' performance and applicability in diverse clinical scenarios.

In conclusion, this paper outlines a significant step towards the practical deployment of AI in clinical pathology, showcasing how computational constraints can be effectively managed without compromising diagnostic accuracy.