Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Exploiting Liver CT scans in Colorectal Carcinoma genomics mutation classification

Daniele Perlo, Luca Berton, Alessia Delpiano, Francesca Menchini, Stefano Tibaldi, Marco Grosso, Paolo Fonio

TL;DR

Non-invasive genomic mutation profiling for CRC liver metastases from CT images is addressed. The authors propose a 2D per-lesion DL pipeline with self-supervised pretraining on external liver-CT datasets (SimCLR) and optional self-attention, evaluated on the CRC4AI dataset. They demonstrate a $F_1$ score of $0.73$ for the RAS mutation family at $128×128$ inputs with OOD pretraining, and show that three-class mutation grouping improves AUC while maintaining balanced sensitivity and specificity. This work lays groundwork for real-time, non-invasive mutation-informed therapy decisions, while acknowledging data scarcity and the need for 3D extensions and larger datasets for broader mutation coverage.

Abstract

The liver is the most involved organ by distant metastasis in colon-rectal cancer (CRC) patients and it comes necessary to be aware of the mutational status of the lesions to correctly design the best individual treatment. So far, efforts have been made in order to develop non-invasive and real-time methods that permit the analysis of the whole tumor, using new artificial intelligence tools to analyze the tumor's image obtained by Computed Tomography (CT) scan. In order to address the current medical workflow, that is biopsy analysis-based, we propose the first DeepLearning-based exploration, to our knowledge, of such classification approach from the patient medical imaging. We propose i) a solid pipeline for managing undersized datasets of available CT scans and ii) a baseline study for genomics mutation diagnosis support for preemptive patient follow-up. Our method is able to identify CRC RAS mutation family from CT images with 0.73 F1 score.

Exploiting Liver CT scans in Colorectal Carcinoma genomics mutation classification

TL;DR

Non-invasive genomic mutation profiling for CRC liver metastases from CT images is addressed. The authors propose a 2D per-lesion DL pipeline with self-supervised pretraining on external liver-CT datasets (SimCLR) and optional self-attention, evaluated on the CRC4AI dataset. They demonstrate a score of for the RAS mutation family at inputs with OOD pretraining, and show that three-class mutation grouping improves AUC while maintaining balanced sensitivity and specificity. This work lays groundwork for real-time, non-invasive mutation-informed therapy decisions, while acknowledging data scarcity and the need for 3D extensions and larger datasets for broader mutation coverage.

Abstract

The liver is the most involved organ by distant metastasis in colon-rectal cancer (CRC) patients and it comes necessary to be aware of the mutational status of the lesions to correctly design the best individual treatment. So far, efforts have been made in order to develop non-invasive and real-time methods that permit the analysis of the whole tumor, using new artificial intelligence tools to analyze the tumor's image obtained by Computed Tomography (CT) scan. In order to address the current medical workflow, that is biopsy analysis-based, we propose the first DeepLearning-based exploration, to our knowledge, of such classification approach from the patient medical imaging. We propose i) a solid pipeline for managing undersized datasets of available CT scans and ii) a baseline study for genomics mutation diagnosis support for preemptive patient follow-up. Our method is able to identify CRC RAS mutation family from CT images with 0.73 F1 score.
Paper Structure (10 sections, 1 equation, 5 figures, 4 tables, 1 algorithm)

This paper contains 10 sections, 1 equation, 5 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Dataset
  4. Targets and Annotations
  5. Method
  6. CT Preprocessing
  7. Self-Supervised Pretraining
  8. Genomics mutation classification
  9. Results
  10. Conclusion

Figures (5)

  • Figure 1: Correlation matrix between lesion mutations. A significant correlation can be find between PiK3CA and BRAF samples.
  • Figure 2: Samples from data distribution: multiple examples of CRC4AI lesions, the last two images in the bottom row are taken form IRCADb and LiTS17 respectively.
  • Figure 3: Comparison of F1 score and AUC between the classifications with 5 and 3 classes. We can notice how OODp+SA is more prone to overfit without any early-stopping policy.
  • Figure 4: Comparison of AUC and Hamming loss(lower is better) of the OODp+SA model with the different resolutions. The small variance between different input resolutions indicates a model preference for coarse morphological features, rather than finer image details.
  • Figure 5: Comparison among the F1 scores of the different models with different image resolutions. In each scenario, a promising RAS classification capability is evident for OODp+SA model.