[1291] A Quantitative Histomorphometric Classifier Identifies Aggressive Versus Indolent p16 Positive Oropharyngeal Squamous Cell Carcinoma

James S Lewis, Jr., Sahirzeeshan Ali, Wade L Thorstad, Anant Madabhushi. Washington University, St. Louis, MO; Case Western Reserve University, Cleveland, OH

Background: Human papillomavirus-related (p16 positive) oropharyngeal squamous cell carcinoma (OSCC) represents a steadily increasing proportion of head and neck cancers and has a favorable prognosis. However, approximately 10% of patients develop recurrent disease, mostly distant metastasis, and the remaining patients often have major morbidity from treatment. Better knowledge about which patients have more aggressive tumors versus more indolent ones is critically important. We recently identified cellular anaplasia and multinucleation as associated with disease recurrence, and the current work sought to find a computer-aided histomorphometric classifier that could detect changes to predict tumor behavior simply from automated digital image analysis of H&E slides.
Design: Using a tissue microarray cohort of p16 positive OSCC cases with clinical follow up, digitally scanned H&E images were marked binarily according to tumor recurrence versus none. Each nucleus was identified via an automated computerized image analysis algorithm. Then, using a novel cluster cell graph that measures the spatial distribution and clustering of cells, a series of topological features defined on each node of the subgraph (i.e., local graph metrics such as clustering coefficients, ratio of connected components and skewness of edge lengths) were analyzed and a random forest decision tree classifier developed and trained. Then, over 25 runs of 3-fold cross validation using subsets of the cases for independent training and testing, the classifier was validated.
Results: There were 160 p16+ patients on the array, of whom 19 (11.9%) developed recurrent disease. Classifier results are listed in Table 1. The classifier was correct in 140 cases (87.5%), had a 47.8% positive predictive value, and 94.2% negative predictive value.

Table 1
 Classifier +Classifier -Total
Recurrence11 (47.8%)8 (5.8%)19
No Recurrence12 (52.2%)129 (94.2%)141

In univariate analysis, patients with a positive classifier had poorer overall, disease free, and disease specific survival (p<0.001 for each).
Conclusions: Based only on tiny H&E punches, a computer-aided morphometric classifier can strongly predict tumors at low likelihood of recurrence. With further validation, this may be a very useful in practice to select patients for de-escalated therapies versus those who should receive more aggressive treatment.
Category: Head & Neck

Wednesday, March 6, 2013 9:30 AM

Poster Session V # 192, Wednesday Morning


Close Window