Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Region of Interest Detection in Melanocytic Skin Tumor Whole Slide Images -- Nevus & Melanoma

Yi Cui, Yao Li, Jayson R. Miedema, Sharon N. Edmiston, Sherif Farag, J. S. Marron, Nancy E. Thomas

TL;DR

Region-of-interest detection in melanocytic skin tumor WSIs is challenging due to limited annotations and inter-pathologist discordance. The authors propose PCLA-3C, a patch-based, three-class classifier that leverages partial ROI annotations to guide ROI detection and ranks patches by class scores to form ROI maps, evaluated with $IoU$ and the annotation ratio $\beta = \frac{A_p}{C_p}$. On the UNC Melanocytic Tumor Dataset, PCLA-3C achieves slide accuracy of $0.923$ and ROI IoU of $0.382$, outperforming the CLAM baseline ($0.692$, $0.112$), with robustness demonstrated across training splits. The approach provides accurate ROI localization and slide classification with limited annotations, enabling faster, cost-effective digital pathology and potential generalization to other cancer WSIs.

Abstract

Automated region of interest detection in histopathological image analysis is a challenging and important topic with tremendous potential impact on clinical practice. The deep-learning methods used in computational pathology may help us to reduce costs and increase the speed and accuracy of cancer diagnosis. We started with the UNC Melanocytic Tumor Dataset cohort that contains 160 hematoxylin and eosin whole-slide images of primary melanomas (86) and nevi (74). We randomly assigned 80% (134) as a training set and built an in-house deep-learning method to allow for classification, at the slide level, of nevi and melanomas. The proposed method performed well on the other 20% (26) test dataset; the accuracy of the slide classification task was 92.3% and our model also performed well in terms of predicting the region of interest annotated by the pathologists, showing excellent performance of our model on melanocytic skin tumors. Even though we tested the experiments on the skin tumor dataset, our work could also be extended to other medical image detection problems to benefit the clinical evaluation and diagnosis of different tumors.

Region of Interest Detection in Melanocytic Skin Tumor Whole Slide Images -- Nevus & Melanoma

TL;DR

Region-of-interest detection in melanocytic skin tumor WSIs is challenging due to limited annotations and inter-pathologist discordance. The authors propose PCLA-3C, a patch-based, three-class classifier that leverages partial ROI annotations to guide ROI detection and ranks patches by class scores to form ROI maps, evaluated with and the annotation ratio . On the UNC Melanocytic Tumor Dataset, PCLA-3C achieves slide accuracy of and ROI IoU of , outperforming the CLAM baseline (, ), with robustness demonstrated across training splits. The approach provides accurate ROI localization and slide classification with limited annotations, enabling faster, cost-effective digital pathology and potential generalization to other cancer WSIs.

Abstract

Automated region of interest detection in histopathological image analysis is a challenging and important topic with tremendous potential impact on clinical practice. The deep-learning methods used in computational pathology may help us to reduce costs and increase the speed and accuracy of cancer diagnosis. We started with the UNC Melanocytic Tumor Dataset cohort that contains 160 hematoxylin and eosin whole-slide images of primary melanomas (86) and nevi (74). We randomly assigned 80% (134) as a training set and built an in-house deep-learning method to allow for classification, at the slide level, of nevi and melanomas. The proposed method performed well on the other 20% (26) test dataset; the accuracy of the slide classification task was 92.3% and our model also performed well in terms of predicting the region of interest annotated by the pathologists, showing excellent performance of our model on melanocytic skin tumors. Even though we tested the experiments on the skin tumor dataset, our work could also be extended to other medical image detection problems to benefit the clinical evaluation and diagnosis of different tumors.
Paper Structure (12 sections, 5 figures, 2 tables)

This paper contains 12 sections, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Data
  4. Data Preparation
  5. Model training and assessment
  6. Results
  7. Method Comparison
  8. Model Validation and Robustness
  9. Misclassified Slides Discussion
  10. Discussion
  11. Acknowledgments.
  12. Disclosure of Interests.

Figures (5)

  • Figure 1: ROI was annotated by black dots determined by pathologists. The predicted ROI was bounded by the green line on the right.
  • Figure 2: Overview of the proposed detection framework. (a) The Melanocytic Tumor Dataset: Randomly assigned 80% (134 WSIs) of data as the training set and 20% (26 WSIs) of data as the testing set. (b) Preprocessing: color normalization Macenko2009ANor and data augmentation. (c) Extract melanoma, nevus and other patches from training data. (d) Model Training: Trained a 3-class patch classifier based on extracted patches. (e) Slide Classification: For each slide, generated predicted scores for all patches and calculated patch as well as slide classification accuracy. (f) Patch Ranking: Ranked all patches from a slide based on the corresponding predicted scores in the context of melanoma or nevus, depending on the slide classification result. (g) Visualization: Generated visualization results based on predicted scores.
  • Figure 3: Visualization results for a melanoma sample and a nevus sample.
  • Figure 4: Visualization results for a misclassified case 1.
  • Figure 5: Visualization results for a misclassified case 2.