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

Xiaoqi Li, Haoyuan Chen+7 more|arXiv (Cornell University)|Jan 21, 2022

AI in cancer detection4 citations

TL;DR

This comprehensive review evaluates machine vision techniques for lymphatic histopathology image analysis, focusing on segmentation, classification, and detection using deep learning. It highlights U-Net as a dominant segmentation method and demonstrates that deep learning features outperform traditional handcrafted features, achieving up to 90.35% accuracy in classifying follicles in follicular lymphoma images.

ABSTRACT

In the past ten years, the computing power of machine vision (MV) has been continuously improved, and image analysis algorithms have developed rapidly. At the same time, histopathological slices can be stored as digital images. Therefore, MV algorithms can provide doctors with diagnostic references. In particular, the continuous improvement of deep learning algorithms has further improved the accuracy of MV in disease detection and diagnosis. This paper reviews the applications of image processing technology based on MV in lymphoma histopathological images in recent years, including segmentation, classification and detection. Finally, the current methods are analyzed, some more potential methods are proposed, and further prospects are made.

Motivation & Objective

• To systematically review machine vision applications in lymphatic histopathology image analysis over the past decade.
• To analyze the evolution of image preprocessing, segmentation, feature extraction, classification, and detection methods in lymphoma diagnosis.
• To identify limitations in current approaches and propose future directions for interpretable, efficient, and accurate deep learning-based systems.
• To evaluate the performance of state-of-the-art methods using benchmark datasets such as Camelyon and IICBU-2008.
• To assess the clinical potential of machine vision in supporting pathologists through computer-aided diagnosis in lymphoma.

Proposed method

• Review of 100+ studies on machine vision in lymphatic histopathology from 2010 to 2022, focusing on image preprocessing, segmentation, feature extraction, classification, and detection.
• Use of U-Net as the dominant deep learning architecture for image segmentation, particularly in lymph node and follicle region detection.
• Application of k-means clustering and color-texture feature extraction (via color space conversion and co-occurrence matrices) for unsupervised segmentation of follicular structures.
• Employment of Hough transform and intensity thresholding for nuclear detection, followed by neuro-fuzzy classification to distinguish centroblasts (CBs) from non-CBs.
• Comparison of deep learning features (e.g., from CNNs) with traditional handcrafted features (color, texture, shape) in classification tasks.
• Evaluation of methods using standard metrics such as segmentation accuracy (e.g., 83.09 ± 6.25%) and classification F1-scores, with datasets including Camelyon and IICBU-2008.

Experimental results

Research questions

• RQ1 How have machine vision techniques evolved in lymphatic histopathology image analysis since 2010?
• RQ2 What are the most effective image preprocessing, segmentation, and feature extraction methods for lymphoma tissue images?
• RQ3 To what extent do deep learning-based features outperform traditional handcrafted features in classifying lymphoma subtypes?
• RQ4 How accurate are current methods in detecting and classifying key histological structures such as follicles and centroblasts?
• RQ5 What are the key challenges and future directions for developing clinically deployable, interpretable, and efficient machine vision systems in lymphoma diagnosis?

Key findings

• U-Net has emerged as the dominant deep learning architecture for image segmentation in lymphatic histopathology, significantly improving accuracy over traditional clustering methods.
• Deep learning features consistently outperform handcrafted features (e.g., color, texture, shape) in classification and detection tasks, especially in complex lymphoma subtypes.
• The neuro-fuzzy classifier achieved 90.35% detection accuracy for centroblasts (CBs) in follicular lymphoma, outperforming SVM, Adaboost, neural networks, and decision trees.
• Segmentation using k-means and color-texture features achieved an average accuracy of 83.09 ± 6.25% on IHC-stained images, demonstrating robustness in follicle region detection.
• The Camelyon and IICBU-2008 datasets are widely used benchmarks for evaluating lymphoma image analysis systems, enabling reproducible performance comparisons.
• Despite progress, challenges remain in interpretability, computational efficiency, and deployment in clinical settings, especially post-COVID-19, where demand for medical AI has surged.