Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Quantifying Cancer Likeness: A Statistical Approach for Pathological Image Diagnosis

Toshiki Kindo

TL;DR

A new statistical approach to automatically identify cancer regions in pathological images that has the practical advantage of not requiring a precise demarcation line between cancer and normal and frees pathologists from the monotonous and tedious work of building consensus with other pathologists.

Abstract

In this paper, we present a new statistical approach to automatically identify cancer regions in pathological images. The proposed method is built from statistical theory in line with evidence-based medicine. The two core technologies are the classification information of image features, which was introduced based on information theory and which cancer features take positive values, normal features take negative values, and the calculation technique for determining their spatial distribution. This method then estimates areas where the classification information content shows a positive value as cancer areas in the pathological image. The method achieves AUCs of 0.95 or higher in cancer classification tasks. In addition, the proposed method has the practical advantage of not requiring a precise demarcation line between cancer and normal. This frees pathologists from the monotonous and tedious work of building consensus with other pathologists.

Quantifying Cancer Likeness: A Statistical Approach for Pathological Image Diagnosis

TL;DR

A new statistical approach to automatically identify cancer regions in pathological images that has the practical advantage of not requiring a precise demarcation line between cancer and normal and frees pathologists from the monotonous and tedious work of building consensus with other pathologists.

Abstract

In this paper, we present a new statistical approach to automatically identify cancer regions in pathological images. The proposed method is built from statistical theory in line with evidence-based medicine. The two core technologies are the classification information of image features, which was introduced based on information theory and which cancer features take positive values, normal features take negative values, and the calculation technique for determining their spatial distribution. This method then estimates areas where the classification information content shows a positive value as cancer areas in the pathological image. The method achieves AUCs of 0.95 or higher in cancer classification tasks. In addition, the proposed method has the practical advantage of not requiring a precise demarcation line between cancer and normal. This frees pathologists from the monotonous and tedious work of building consensus with other pathologists.
Paper Structure (9 sections, 12 equations, 3 figures, 1 table)

This paper contains 9 sections, 12 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Statistical theory of Useful Image Features
  3. Kullback-Leibler Divergence of Feature Distribution on Classification Space
  4. Spatial Distribution of Classification Information Content
  5. Experiments using CAMELYON16 data
  6. Probability Estimation within the Classification Space
  7. Rapid Learning Algorithm and Results
  8. Short Discussion
  9. Concluding Remark

Figures (3)

  • Figure 1: A schematic figure shows the outline of the method proposed. The classification information content $C_{KL}(\rho^p(f)||\rho^p(f^0))$ is simply represented as $C_{KL}(\rho^p(f))$.
  • Figure 2: The image on the far left displays the annotation image of tumor_011 in the CAMELYON16 dataset. The subsequent figure shows a total of 20 cancer patches outlined in blue and 20 normal patches outlined in red, all of which were selected from tumor_011 in the CAMELYON16 dataset. The next image is the classification results for tumor_011 are depicted, with cancerous areas marked in blue and normal areas in red. Moving on, the outcome of classifying another tumor, tumor_009-004, using patches selected from tumor_011 is presented with its annotation image.
  • Figure 3: A histogram showing the number of image patches(blue-cancer and red-normal), with the classification information content of the image patches on the horizontal axis. The vertical red line indicates the position where the classification information content is zero.