Utilizing Artificial Intelligence for Head and Neck Cancer Outcomes Prediction From Imaging

Abstract

Artificial intelligence (AI)-based models have become a growing area of interest in predictive medicine and have the potential to aid physician decision-making to improve patient outcomes. Imaging and radiomics play an increasingly important role in these models. This review summarizes recent developments in the field of radiomics for AI in head and neck cancer. Prediction models for oncologic outcomes, treatment toxicity, and pathological findings have all been created. Exploratory studies are promising; however, validation studies that demonstrate consistency, reproducibility, and prognostic impact remain uncommon. Prospective clinical trials with standardized procedures are required for clinical translation.

Progression Free Survival Prediction for Head and Neck Cancer using Deep Learning based on Clinical and PET-CT Imaging Data

“…Interestingly, the addition of clinical data generally did not make any noticeable improvements in model performance, regardless of the C-index metric calculation method, possibly indicating the majority of informative data was contained within the PET/CT images. This observation runs counter to results seen in other clinical prediction models, where the addition of clinical data to imaging information often improves performance [6].…”

“…The determination of prognostic outcomes is an unmet need for HNSCC patients that could improve clinical decision-making processes. While the performance of our models (as measured in C-index without events observed) is not ideal, they are reasonable within the context of prognostic prediction, which is known to be notoriously complex task [6]. Our main innovation stems from the use of ensembling approaches applied to predictions from internal cross-validation, which seems to reasonably improve overall performance on unseen data.…”

“…Therefore, PET/CT imaging data has been as an attractive biomarker in developing clinical outcome prediction models for HNSCC. Radiomic analysis of PET/CT images in HNSCC has been heavily investigated in the past, with models based on statistical, geometric, and texture information contained within regions of interest showing promise in predicting PFS [5,6]. However, radiomic methods rely on hand-crafted features and pre-defined regions of interest, which can introduce bias into analysis pipelines [7].…”

A winning combination in the war against deadly cancers: exploring new opportunities for combining antibody–drug conjugates and cancer immunotherapies

“…Machine learning was used to create a model for predicting treatment in oropharyngeal squamous cell carcinoma while taking into consideration variables related to the tumors, socioeconomic, regional, and institutional factors [52]. Artificial intelligence is also used for Intensity-modulated radiation therapy (IMRT) treatment planning [53] and radiomics in head and neck cancer [54]. Furthermore, artificial intelligence was used to predict microsatellite instability and deficient DNA mismatch repair in hematoxylin and eosin stained colorectal cancer sections with high accuracy in uniform datasets [55].…”

Open-source microscopic solution for classification of biological samples

“…Artificial intelligence and machine learning have also found applications in cyber security, 31 autonomous driving 32 and cancer research. 33 Google Brain built Tensorflow as an API for implementing and executing machine learning algorithms. TensorFlow supports both C++ and Python as front-end languages.…”