Quantitative analysis of liver tumors at different stages using microscopic hyperspectral imaging technology

Jiansheng Wang, Qingli LiJournal of Biomedical Optics 2018 · DOI

TL;DR

This paper introduces a novel method for quantitatively analyzing liver tumors at different stages using microscopic hyperspectral imaging (MHSI). By combining morphological watershed algorithms for segmentation and support vector machines for classification, the approach accurately quantifies tumor area and shows promise for early liver cancer diagnosis.

What Problem Does This Paper Address?

Liver cancer is a significant global health concern with a low survival rate, often due to late diagnosis. Traditional pathological diagnosis relies on visual inspection of microscopic liver sections, which can be subjective and may miss early signs of malignancy. While various imaging techniques exist, they often have limitations in early detection or are invasive. This paper aims to address the challenge of early and accurate liver tumor diagnosis by developing a quantitative analysis method using MHSI, which captures both spatial and spectral information from tissue samples. The goal is to provide a more objective and sensitive approach to identifying and quantifying liver tumors at various stages of development.

Methods

The proposed method involves a multi-step process:

  • Image Acquisition: Microscopic hyperspectral images of rat liver tissue samples were acquired using an acousto-optic tunable filter (AOTF)-based MHSI platform. This system captures images across a spectral range of 550-1000 nm with a spectral resolution of 2-6 nm.
  • Image Preprocessing:
    • Spectral Domain Normalization: This step calibrates spectral responses to obtain relative transmittance, eliminating image artifacts and noise caused by the light source and tissue preparation.
  • Spatial Domain Enhancement: Median filtering is applied to remove salt-and-pepper noise and improve the visual quality of the images.
  • Spatial Analysis: Grayscale Image Segmentation: A morphological watershed algorithm (MWA) is employed to segment the hyperspectral images. This algorithm leverages spatial information to delineate different regions within the tissue, which is crucial for identifying tumor boundaries, especially in early stages where visual differences are subtle.
  • Spectral Analysis: Liver Tumor Identification: A supervised classification algorithm, specifically a Support Vector Machine (SVM) with a Gaussian kernel, is used to classify the segmented regions. The SVM utilizes both spectral and spatial-textural information extracted from the hyperspectral data to distinguish between liver tumors and normal tissue.
  • Postclassification and Evaluation:
    • Majority/Minority Analysis (MMA): This post-classification step is used to eliminate small, misclassified regions (plaques) that may resemble tumor tissue but are not.
  • Quantitative Analysis: The percentage of tumor area in the liver samples is calculated based on the classified pixels.
  • Pseudocolor Synthesis: To enhance visualization, pseudocolor images are generated by combining selected spectral bands. This technique aims to highlight tumor morphology and distribution for better visual interpretation.
  • Evaluation Metrics: Classification performance is assessed using Overall Accuracy (OA) and the Kappa coefficient, comparing the results against standard classification images provided by pathologists.

Datasets

The study utilized in vitro samples of rat models of bile duct carcinoma. The experimental group consisted of 40 SD male rats induced with thioacetamide (TAA) to develop liver tumors. Control groups of rats were also maintained. Liver tissue samples were collected at different time points: 8, 12, 16, and 20 weeks post-TAA administration. These samples were then processed into pathological sections for MHSI acquisition. The paper does not introduce a new benchmark dataset but rather uses these experimental samples for analysis and comparison with pathologist-provided standard classification images.

Key Insights and Ideas

  • MHSI for Early Diagnosis: The study demonstrates that MHSI, by capturing rich spectral information, can reveal subtle differences between normal and cancerous liver tissue even at early stages (e.g., 8 weeks), which might be missed by conventional microscopy.
  • Synergy of MWA and SVM: The combination of MWA for robust spatial segmentation and SVM for spectral classification proves effective in accurately identifying and delineating liver tumors. The MWA helps overcome the limitations of purely spectral methods by incorporating spatial context.
  • Quantitative Tumor Area as a Biomarker: The ability to accurately quantify the percentage of tumor area in liver biopsies provides a valuable quantitative metric for assessing tumor progression and potentially for early diagnosis.
  • Pseudocolor Synthesis for Enhanced Visualization: The pseudocolor synthesis technique effectively highlights tumor regions, improving visual contrast and making it easier to interpret pathological changes compared to standard true-color images.

Key Results

  • Quantitative Tumor Area: The calculated percent tumor areas at different time points were: 3.338% at 8 weeks, 11.952% at 12 weeks, 15.125% at 16 weeks, and 23.375% at 20 weeks. This shows a clear trend of increasing tumor area with disease progression.
  • Classification Accuracy: The proposed MWA-based SVM method achieved high classification accuracy, with Kappa coefficients consistently above 0.8 (ranging from 0.8257 to 0.9197) across the different time points. This indicates strong classification consistency. In comparison, the standard SVM method showed significantly lower Kappa coefficients (generally <0.6), highlighting the advantage of the proposed approach.
  • Early Diagnosis Potential: The analysis of 8-week images demonstrated the potential for early diagnosis, with the method being able to identify minor lesions.
  • Visual Enhancement: Pseudocolor composite images effectively visualized tumor morphology and distribution, offering enhanced contrast and clarity compared to original true-color images.

Limitations and Future Work

  • Limitations:
    • The study was conducted on in vitro rat models, and further validation on human samples is necessary.
  • While the method shows promise for early diagnosis, the "no tumor components" in control groups and the variability in tumor area within the same time point (especially at later stages) suggest that further refinement might be needed for absolute certainty.
  • The current work does not deeply refine the identification of specific liver tumor components beyond distinguishing tumor from normal tissue.
  • Future Work:
    • The authors suggest using more liver tissue sections from the early stages (8 weeks and earlier) to further verify the method's validity.
  • Future research should focus on extracting more complex liver tissue features to achieve more accurate information for early diagnosis.
  • Expanding the analysis to include a wider range of liver tumor types and stages would be beneficial.

Summary updated

Abstract

Liver cancer has one of the highest rates of human morbidity and mortality. However, in terms of pathology, liver cancer is traditionally clinically diagnosed based on observation of microscopic images of pathological liver sections. This paper investigates in vitro samples of rat models of bile duct carcinoma and presents a quantitative analysis method based on microscopic hyperspectral imaging technology to evaluate liver cancers at different stages. The example-based feature extraction method used in this paper mainly includes two algorithms: a morphological watershed algorithm is applied to find object and segment pathological components of pathological liver sections at different stages, and a support vector machine algorithm is implemented for liver tumor classification. Majority/minority analysis is utilized as the postclassification tool to eliminate small plaques from the preliminary classification results. Then, pseudocolor synthesis in RGB color space is used to produce the final results. The experimental results show that this method can effectively calculate the percent tumor areas in liver biopsies at different time points, that is, 3.338%, 11.952%, 15.125%, and 23.375% at 8, 12, 16, and 20 weeks, respectively. Notably, through tracking analysis, the processed results of 8-week images showed the possibility for early diagnosis of the liver tumor.
BibTeX
@article{wang2018quantitative,
  author = {Jiansheng Wang and Qingli Li},
  title = {{Quantitative analysis of liver tumors at different stages using microscopic hyperspectral imaging technology}},
  journal = {Journal of Biomedical Optics},
  year = {2018},
  doi = {10.1117/1.jbo.23.10.106002},
}
Figure