Automatic Prediction of Cancer Recurrence through Machine Learning

SUMMARY

• Deep learning tools are becoming increasingly popular for histologic analyses and have been used to predict tumor biomarker status, clinical variables, tumor subtypes, and mutation status in a variety of cancers.
• Accurate prediction of recurrence risk for post-operative cancer recurrence is of significant clinical interest as it can be used to recommend adjuvant therapy for those highest risk individuals.
• The faculty trained a deep convolutional neural network (DCNN) to identify patterns in digital images of standard hematoxylin and eosin stained diagnostic pathology slides from respected HPV+ oropharynx head/neck cancer patients.
• Proper identification of these highest risk individuals is critically important because adjuvant therapy can come with significant long-term toxicity.

FIGURE

ADVANTAGES

ADVANTAGES

• Enables identification of key morphological features in hematoxylin and eosin stained diagnostic pathology slides from resected HPV+ oropharynx head/neck cancer patients.
• Algorithm builds on HPV prediction and includes clinical recurrence prediction

APPLICATIONS

• Diagnostics
• Improved workflow in processing images from routine hematoxylin and eosin (HE) stained slides
• Identification of patients at highest risk of post-operative HPV+ cancer recurrence
• Fast, low-cost method to identify virus-induced cancer recurrence in clinical trials or clinical routine
• Feature visualization for pathologists to verify plausibility of computer-based image classification

PUBLICATIONS

• Dolezal, J.M., Trzcinska, A., Liao, CY. et al. Deep learning prediction of BRAF-RAS gene expression signature identifies noninvasive follicular thyroid neoplasms with papillary-like nuclear features. Mod Pathol 34, 862–874 (2021).