Deep learning in head & neck cancer outcome prediction

Abstract

Traditional radiomics involves the extraction of quantitative texture features from medical images in an attempt to determine correlations with clinical endpoints. We hypothesize that convolutional neural networks (CNNs) could enhance the performance of traditional radiomics, by detecting image patterns that may not be covered by a traditional radiomic framework. We test this hypothesis by training a CNN to predict treatment outcomes of patients with head and neck squamous cell carcinoma, based solely on their pre-treatment computed tomography image. The training (194 patients) and validation sets (106 patients), which are mutually independent and include 4 institutions, come from The Cancer Imaging Archive. When compared to a traditional radiomic framework applied to the same patient cohort, our method results in a AUC of 0.88 in predicting distant metastasis. When combining our model with the previous model, the AUC improves to 0.92. Our framework yields models that are shown to explicitly recognize traditional radiomic features, be directly visualized and perform accurate outcome prediction.

Multi-Level Multi-Modality Fusion Radiomics: Application to PET and CT Imaging for Prognostication of Head and Neck Cancer

“…In the study by Diamant et al [34], deep learning was only applied to CT images while clinical parameters were not integrated, and only AUC values were reported (requiring a relatively ad hoc time-to-event cut-off threshold), while the C-index (which preserves the continuous time-to-outcome information) was not reported. Thus, our results are not directly comparable; but to summarize, the authors obtained AUC values of 0.65, 0.88 and 0.70 for RFS, MFS and OS, in comparison to our C-index values of 0.67, 0.82 and 0.70.…”

List of Deep Learning Models

“…There has been an enormous evolution in system modeling and intelligence after introducing the early models for deep learning [1][2][3][4][5][6][7][8]. Deep learning methods very fast emerged and expanded applications in various scientific and engineering domains.…”

Integrating deep and radiomics features in cancer bioimaging

“…This work aimed mainly at investigating a new framework for the integration of radiomics and deep learning; limited effort was focused on tuning the DL model, and we restricted the types of radiomics features to those proposed in the reference paper [38]. A similar combination approach of deep and radiomic features has been applied on a subset of the HN dataset to predict distant metastasis by applying CNNs to CT scans only [40]. In particular, we expect that better accuracy can be achieved by adopting specific deep learning architectures or considering more complex methods to extract radiomics features, for instance applying Wavelet filters [17].…”

“…The Head-Neck-PET-CT (HN) dataset 1 has been originally introduced in [38], and further used in [40]. It includes medical images and clinical data of 298 patients with head and neck squamous cell carcinoma.…”