TopOC: Topological Deep Learning for Ovarian and Breast Cancer Diagnosis

TL;DR

This paper proposes the integration of topological deep learning methods to enhance the accuracy and robustness of existing histopathological image analysis models and demonstrates that the inclusion of topological features significantly improves the differentiation of tumor types in ovarian and breast cancers.

Abstract

Microscopic examination of slides prepared from tissue samples is the primary tool for detecting and classifying cancerous lesions, a process that is time-consuming and requires the expertise of experienced pathologists. Recent advances in deep learning methods hold significant potential to enhance medical diagnostics and treatment planning by improving accuracy, reproducibility, and speed, thereby reducing clinicians' workloads and turnaround times. However, the necessity for vast amounts of labeled data to train these models remains a major obstacle to the development of effective clinical decision support systems. In this paper, we propose the integration of topological deep learning methods to enhance the accuracy and robustness of existing histopathological image analysis models. Topological data analysis (TDA) offers a unique approach by extracting essential information through the evaluation of topological patterns across different color channels. While deep learning methods capture local information from images, TDA features provide complementary global features. Our experiments on publicly available histopathological datasets demonstrate that the inclusion of topological features significantly improves the differentiation of tumor types in ovarian and breast cancers.

Figures (5)

• Figure 1: For the $5\times 5$ image $\mathcal{X}$ with the given pixel values, the sublevel filtration is the sequence of binary images $\mathcal{X}_1\subset \mathcal{X}_2\subset \mathcal{X}_3\subset \mathcal{X}_4\subset \mathcal{X}_5$.
• Figure 2: TopOC-1 Model. We first generate persistence diagrams for any input image, utilizing grayscale values. Next, we derive our topological feature vectors, represented as Betti functions. These vectors are then inputted into a simple ML classifier to produce classification results.
• Figure 3: TopOC-CNN Model. In this model, we integrate topological features of images with convolutional vectors from a CNN backbone, followed by a fully connected layer. For the prediction head, we concatenate (a) a 128-dimensional CNN vector and (b) a topological vector, with dimension options of 256, 128, 64, or 0 (Vanilla-CNN).
• Figure 4: In the left figure, we present the median curves and 40% confidence bands of Betti-0 (grayscale) curves for five classes from the UBC-OCEAN dataset. In the next two figures, we display the same curves for two specific classes.
• Figure 5: Median curves and $40\%$ confidence bands of our topological feature vectors for each class for BREAKHIS dataset.