Abstract
Cancer cell lines are widely used for research purposes in laboratories all over the world. Computer-assisted classification of cancer cells can alleviate the burden of manual labeling and help cancer research. In this paper, we present a novel computerized method for cancer cell line image classification. The aim is to automatically classify 14 different classes of cell lines including 7 classes of breast and 7 classes of liver cancer cells. Microscopic images containing irregular carcinoma cell patterns are represented by subwindows which correspond to foreground pixels. For each subwindow, a covariance descriptor utilizing the dual-tree complex wavelet transform (DT-$ℂ$WT) coefficients and several morphological attributes are computed. Directionally selective DT-$ℂ$WT feature parameters are preferred primarily because of their ability to characterize edges at multiple orientations which is the characteristic feature of carcinoma cell line images. A Support Vector Machine (SVM) classifier with radial basis function (RBF) kernel is employed for final classification. Over a dataset of 840 images, we achieve an accuracy above 98%, which outperforms the classical covariance-based methods. The proposed system can be used as a reliable decision maker for laboratory studies.
Introduction
In cancer research, the use of authenticated cell lines is critical for reproducible results. However, manual identification is labor-intensive and prone to error, leading to frequent misidentification or cross-contamination. While Short Tandem Repeat (STR) analysis remains the gold standard for authentication, its high cost and time requirements limit its routine use. This study proposes an automated, image-based classification system that identifies cell lines based on their unique morphological signatures in microscopic images, providing a cost-effective alternative to genetic fingerprinting.
Methodology
The proposed classification pipeline involves several sophisticated image processing steps designed to capture the irregular patterns of carcinoma cells.
Segmentation and Pre-processing
To ensure the classifier focuses on biological data, the system separates foreground cells from the background using an Expectation-Maximization (EM) algorithm.
Feature Extraction via DT-$ℂ$WT
The core of the feature extraction process is the Dual-Tree Complex Wavelet Transform (DT-$ℂ$WT). Unlike standard wavelets, DT-$ℂ$WT is directionally selective, capturing edges and ridges at six distinct orientations. This is crucial for carcinoma cells, which often exhibit irregular, directional growth patterns.
Covariance Descriptors and Classification
Rather than using raw feature vectors, the system calculates how various attributes—such as intensity, derivatives, and wavelet magnitudes—vary together within small sub-windows. These Normalised Covariance matrices provide a robust representation resistant to lighting changes. Finally, a Support Vector Machine (SVM) with a Radial Basis Function (RBF) kernel performs the 14-way classification.
Experimental Results
The system was evaluated using a dataset of 840 microscopic images covering 14 cell lines (7 breast and 7 liver cancer types), with 60 images per class.
Classification Performance
The proposed method achieved an average accuracy exceeding 98%. This performance significantly outperforms traditional methods like the Scale-Invariant Feature Transform (SIFT), which struggled as the number of cell line classes increased. The high accuracy was maintained across multiple magnifications (10x, 20x, and 40x), with 20x magnification identified as providing the optimal balance of cellular detail and field-of-view.
Discussion and Conclusion
The success of this approach lies in its ability to capture the directional "grain" of cancer growth through complex wavelets and the relational information encoded in covariance matrices. While some challenges remain in distinguishing "poorly differentiated" cell lines with highly variable morphologies (such as Mahlavu and SkHep1), the system offers a reliable, cost-effective alternative to genetic authentication. Future work will focus on optimizing computational efficiency and expanding the dataset to further refine classification performance.
Summary updated