OKEN: A Supervised Evolutionary Optimizable Dimensionality Reduction Framework for Whole Slide Image Classification

Abstract

Classification of lung cancer subtypes is a critical clinical step; however, relying solely on H&E-stained histopathology images can pose challenges, and additional immunohistochemical analysis is sometimes required for definitive subtyping. Digital pathology facilitates the use of artificial intelligence for automatic classification of digital tissue slides. Automatic classification of Whole Slide Images (WSIs) typically involves extracting features from patches obtained from them. The aim of this study was to develop a WSI classification framework utilizing an optimizable kernel to encode features from each patch from a WSI into a desirable and adjustable latent space using an evolutionary algorithm. The encoded data can then be used for classification and segmentation while being computationally more efficient. Our proposed framework is compared with a state-of-the-art model, Vim4Path, on an internal and external dataset of lung cancer WSIs. The proposed model outperforms Vim-S16 in accuracy and F₁ score at both ×2.5 and ×10 magnification levels on the internal test set, with the highest accuracy (0.833) and F₁ score (0.721) at ×2.5. On the external test set, Vim-S16 at ×10 achieves the highest accuracy (0.732), whereas OKEN-DenseNet121 at ×2.5 has the best F₁ score (0.772). In future work, finding a dynamic way to tune the output dimensions of the evolutionary algorithm would be of value.

Keywords: deep learning; digital pathology; dimensionality reduction; evolutionary algorithm; lung cancer.

Figure 1

Visualization of data split and distribution. The diagram was created using Mermaid [37]. Abbreviations: HULC: Haukeland University Hospital Lung Cancer; WSIs: whole slide images; TCGA: The Cancer Genome Atlas.

Figure 2

Overview of the proposed three-stage pipeline for tumor region processing and classification. Stage 1: Patch positions are created from the whole slide image (WSI), and a tumor mask is generated using a pretrained model. Tumor regions are identified, and patch positions and tumor patches are saved for downstream processing. Stage 2: An evolutionary algorithm (EA) is used to optimize a transformation matrix by iteratively clustering feature embeddings across generations, leading to better separation between the two classes of labeled data. The final transformation matrices are saved. Stage 3: Features are extracted from the saved patches, followed by dimensionality reduction using the saved transformation matrices and graph formation based on spatial positions. A 1-Dimensional Convolutional Neural Network (1D-CNN) is then applied for classification.

Figure 3

Comparison of dimensionality reduction techniques applied to high-dimensional features extracted from image patches of 22 whole slide images. The (top-left) plot shows the result of the proposed EA-based transformation. The remaining plots display the outputs of UMAP (top-middle and top-right) and t-SNE (bottom-left and bottom-right) applied to both the original features and the EA-based transformed features. The two visualized colors indicate the two tumor subtype classes. Abbreviations: EA: evolutionary algorithm; UMAP: Uniform manifold approximation and projection; t-SNE: t-distributed stochastic neighbor embedding.