Automatic prediction of tumour malignancy in breast cancer with fractal dimension

Abstract

Breast cancer is one of the most prevalent types of cancer today in women. The main avenue of diagnosis is through manual examination of histopathology tissue slides. Such a process is often subjective and error-ridden, suffering from both inter- and intraobserver variability. Our objective is to develop an automatic algorithm for analysing histopathology slides free of human subjectivity. Here, we calculate the fractal dimension of images of numerous breast cancer slides, at magnifications of 40×, 100×, 200× and 400×. Using machine learning, specifically, the support vector machine (SVM) method, the F1 score for classification accuracy of the 40× slides was found to be 0.979. Multiclass classification on the 40× slides yielded an accuracy of 0.556. A reduction of the size and scope of the SVM training set gave an average F1 score of 0.964. Taken together, these results show great promise in the use of fractal dimension to predict tumour malignancy.

Fractal dimension, occupancy and hotspot analyses of B cell spatial distribution predict clinical outcome in breast cancer

“…Fractal dimensions (19) have been used to characterize the irregular morphology of tumors (20)(21)(22) and vasculature (23,24) as well as subcellular structures such as mitochondria (25) and nuclei (26). There are numerous ways to calculate fractal dimensions.…”

“…Previous efforts to quantify the spatial heterogeneity of the tumor microenvironment include measuring the spatial colocalization of tumor and immune cells (13,14), locating immune cell clusters or "hotspots" (15,16), determining the amount of infiltration of lymphocytes into a tumor (17) and using Shannon entropy to quantify the cellular diversity (18). In addition, fractal dimensions (19) have been used to characterize the irregular morphology of tumors (20)(21)(22) and tumor-related structures (23)(24)(25)(26). (We review these approaches in more detail in the Discussion section.…”

Breast Cancer Histopathological Image Classification: A Deep Learning Approach

“…We found that deep ResNet models were more sensitive and reliable than Inception in all tested cancer data-sets. We combined different magnification including 40X, 100X, 200X and 400X to generate comprehensive, independent and scalable system while a large number of previous studies employed single magnification level ( [7], [30]). Several other studies ( [7], [25], [48], [4]) also investigated multiple magnifications of medical images.…”

“…We combined different magnification including 40X, 100X, 200X and 400X to generate comprehensive, independent and scalable system while a large number of previous studies employed single magnification level ( [7], [30]). Several other studies ( [7], [25], [48], [4]) also investigated multiple magnifications of medical images. However, these approaches examined different classifiers for each magnification level and also had medical laboratory limitations to capture required multiple magnification to gather image training samples.…”

“…In this study, we developed and introduced an accurate and reliable computer-based techniques empowered with deep learning approaches to classify cancer types and breast cancer (7,452) phyllodes-tumor (1,362), tubular-adenoma (1,710) sub-types from histopathological images derived from Hematoxylin and eosin stain (H&E) and Immunohistochemistry (IHC) slides. Our framework contains five steps: a) Image acquisition and conversion to JPEG/RGB channels.…”