Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Resource-Limited Automated Ki67 Index Estimation in Breast Cancer

J. Gliozzo, G. Marinò, A. Bonometti, M. Frasca, D. Malchiodi

TL;DR

This paper addresses automated Ki67 and TIL scoring in breast cancer under resource constraints by proposing a resource consumption–aware compression pipeline for a pre-trained PathoNet model. It combines lossy weight sharing quantization with a lossless index-map storage format, enabling inference directly on compressed weights and achieving substantial RAM (~4x) and disk (~9x) reductions, with energy savings around 1.5x and preserved or improved prognostic accuracy. The authors validate the approach on the SHIDC-B-Ki-67-V1.0 dataset, showing comparable Ki67 scores and enhanced TIL scores, while providing a public Python framework to compress, retrain, and deploy models in low-resource settings. The work offers a practical, generalizable strategy for downsizing deep models in histopathology and related medical imaging tasks, with potential extensions to federated learning and other architectures.

Abstract

The prediction of tumor progression and chemotherapy response has been recently tackled exploiting Tumor Infiltrating Lymphocytes (TILs) and the nuclear protein Ki67 as prognostic factors. Recently, deep neural networks (DNNs) have been shown to achieve top results in estimating Ki67 expression and simultaneous determination of intratumoral TILs score in breast cancer cells. However, in the last ten years the extraordinary progress induced by deep models proliferated at least as much as their resource demand. The exorbitant computational costs required to query (and in some cases also to store) a deep model represent a strong limitation in resource-limited contexts, like that of IoT-based applications to support healthcare personnel. To this end, we propose a resource consumption-aware DNN for the effective estimate of the percentage of Ki67-positive cells in breast cancer screenings. Our approach reduced up to 75% and 89% the usage of memory and disk space respectively, up to 1.5x the energy consumption, and preserved or improved the overall accuracy of a benchmark state-of-the-art solution. Encouraged by such positive results, we developed and structured the adopted framework so as to allow its general purpose usage, along with a public software repository to support its usage.

Resource-Limited Automated Ki67 Index Estimation in Breast Cancer

TL;DR

This paper addresses automated Ki67 and TIL scoring in breast cancer under resource constraints by proposing a resource consumption–aware compression pipeline for a pre-trained PathoNet model. It combines lossy weight sharing quantization with a lossless index-map storage format, enabling inference directly on compressed weights and achieving substantial RAM (~4x) and disk (~9x) reductions, with energy savings around 1.5x and preserved or improved prognostic accuracy. The authors validate the approach on the SHIDC-B-Ki-67-V1.0 dataset, showing comparable Ki67 scores and enhanced TIL scores, while providing a public Python framework to compress, retrain, and deploy models in low-resource settings. The work offers a practical, generalizable strategy for downsizing deep models in histopathology and related medical imaging tasks, with potential extensions to federated learning and other architectures.

Abstract

The prediction of tumor progression and chemotherapy response has been recently tackled exploiting Tumor Infiltrating Lymphocytes (TILs) and the nuclear protein Ki67 as prognostic factors. Recently, deep neural networks (DNNs) have been shown to achieve top results in estimating Ki67 expression and simultaneous determination of intratumoral TILs score in breast cancer cells. However, in the last ten years the extraordinary progress induced by deep models proliferated at least as much as their resource demand. The exorbitant computational costs required to query (and in some cases also to store) a deep model represent a strong limitation in resource-limited contexts, like that of IoT-based applications to support healthcare personnel. To this end, we propose a resource consumption-aware DNN for the effective estimate of the percentage of Ki67-positive cells in breast cancer screenings. Our approach reduced up to 75% and 89% the usage of memory and disk space respectively, up to 1.5x the energy consumption, and preserved or improved the overall accuracy of a benchmark state-of-the-art solution. Encouraged by such positive results, we developed and structured the adopted framework so as to allow its general purpose usage, along with a public software repository to support its usage.
Paper Structure (18 sections, 1 equation, 1 figure, 3 tables)

This paper contains 18 sections, 1 equation, 1 figure, 3 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Problem Definition and Data
  4. Algorithm
  5. Notation and preliminary definitions
  6. Compression Framework
  7. The PathoNet model
  8. Lossy compression of network layers
  9. Lossless storage of the compressed network
  10. Model inference in the resource-saver format
  11. Implementation
  12. Results
  13. Evaluation Metrics
  14. Hyperparameters optimization
  15. Experimental results.
  16. ...and 3 more sections

Figures (1)

  • Figure 1: Disk compression ratio of the tested methods with respect to the original size of PathoNet (12.802 MBytes).