Computational protocol: Novel histopathologic feature identified through image analysis augments stage II colorectal cancer clinical reporting

Similar protocols

Protocol publication

[…] The unprocessed, continuous data for each candidate histopathologic feature (tumor budding, poorly differentiated clusters, lymphatic vessel invasion, minimal lymphatic vessel invasion, tumor to stroma ratio and lymphatic vessel density) alongside patient outcome data was loaded into X-Tile software [] to calculate the optimal cut-offs to categorize patients into high or low risk of disease-specific death. The significance of these cut-offs were corrected by cross-validation within Monte-Carlo simulations (n = 1000). Training set cut-offs were subsequently applied to the validation set to categorize the stage II patient population. [...] Principle Component Analysis and Cox regression calculations (univariate and multivariate backwards elimination with a stopping rule based on Aikaike's Information Criteria) were performed in SPSS (IBM, New York, USA). From the 123 multi-parametric feature set the most informative features which differentiated between binary disease specific survival were identified by inputting the continuous data for each parameter into a random forest (n = 5000) decision tree model [] and exporting the associated Gini score. The continuous data from each informative parameter was subsequently input into a classification and regression tree (CART) [] decision tree strategy to identify optimal combinations of and clinical cut-offs of novel significant histopathologic features (Salford Predictive Miner, Salford Systems, San Diego, USA). To avoid over-fitting within the decision tree modelling, validation was performed during the decision tree modelling for both random forest (out of bag) and CART (10 fold self-test). Kaplan Meier curves and associated Benjamini-Hochberg false discovery rate corrected p values were calculated using TMA Navigator ( []. […]

Pipeline specifications

Software tools X-tile, TMA Navigator
Application Tissue array analysis
Organisms Homo sapiens
Diseases Colorectal Neoplasms