Automated tumour budding quantification by machine learning augments TNM staging in muscle-invasive bladder cancer prognosis

Abstract

Tumour budding has been described as an independent prognostic feature in several tumour types. We report for the first time the relationship between tumour budding and survival evaluated in patients with muscle invasive bladder cancer. A machine learning-based methodology was applied to accurately quantify tumour buds across immunofluorescence labelled whole slide images from 100 muscle invasive bladder cancer patients. Furthermore, tumour budding was found to be correlated to TNM (p = 0.00089) and pT (p = 0.0078) staging. A novel classification and regression tree model was constructed to stratify all stage II, III, and IV patients into three new staging criteria based on disease specific survival. For the stratification of non-metastatic patients into high or low risk of disease specific death, our decision tree model reported that tumour budding was the most significant feature (HR = 2.59, p = 0.0091), and no clinical feature was utilised to categorise these patients. Our findings demonstrate that tumour budding, quantified using automated image analysis provides prognostic value for muscle invasive bladder cancer patients and a better model fit than TNM staging.

Identification of Immunological Features Enables Survival Prediction of Muscle-Invasive Bladder Cancer Patients Using Machine Learning

“…In particular, we found that high content of TBs in the invasive frontin, frontout, and tumour core as well as low number of CD68 + cells and high PD-L1 expression in the invasive frontout are indicators of bad prognosis. This supports previous findings in the literature [33,36,66,53]. High density of CD8 + , CD3 + , and CD68 + cells in the invasive frontin, frontout, and tumour core was associated with good prognosis by our models, as similarly reported in other studies [23,67].…”

“…For the quantification of tumours buds, segmentation of the epithelial cells was required. The CNN-RF methodology described by Brieu et al [71] and extended in [36] was adopted. Briefly, a convolutional neural network (CNN) was trained on an annotated data set of epithelium and non-epithelium IF images.…”

“…Finer-grained segmentation masks were produced by the RF, enabling a more accurate segmentation of the epithelium. As detailed in previous work [36], the output of the CNN-RF is finally ensembled with the output of a semantic segmentation network [72] to generate the final epithelium segmentation results. Tumour buds were classified as epithelium objects containing one to four nuclei [36].…”

“…As detailed in previous work [36], the output of the CNN-RF is finally ensembled with the output of a semantic segmentation network [72] to generate the final epithelium segmentation results. Tumour buds were classified as epithelium objects containing one to four nuclei [36]. A total number of 97262 patches were used for training and 9742 patches for validation of the CNN-RF model.…”

“…However, without reliable quantitative methods, tumour budding quantification becomes challenging due to poor inter-observer consistency. [36,37].…”