Integrating visual patterns and textual data into analyzing MOHCCN skin cancer genomics data (melanoma and basal cell carcinoma)

Skin cancers are the most common human cancers in Canada, with at least 89,000 diagnosed cases per year. The sizable number of cases is not only seen in Canada, but several EU countries have also reported that a skin cancer epidemic exists which will most likely accelerate over time, considering the rapid increase in incidences as well as demographic change toward women.

This motivates the development of precision medicine systems for prediction of skin cancer, customized to patient’s profile. In the past decade, machine learning (ML) models, particularly deep neural models, have shown promising performances in image classification and language understanding for precision oncology. ML models have been widely applied to histopathology images for cancer classification and clinical textual reports for automatic extraction of cancer stage for treatment planning. Most of the current efforts, however, focus on a single modality, utilizing only images or textual data.

In this research we propose to integrate visual patterns and textual data (multiple modalities) as auxiliary modalities into genomics data for identifying melanoma and basal cell carcinoma. Different modalities can inform each other, allowing to deal with missing/imperfect data, alleviating the need for labeled data, and improving the overall performances in a precision oncology setting.

Quotes

“With a primary focus on medical text processing, I possess specialized expertise in medical informatics. I am genuinely enthusiastic about the potential of integrating textual, image, and genomics data for predicting the clinical behavior of Basal Cell Carcinoma. Although my exposure to genomics data has been somewhat limited, receiving the esteemed Marathon of Hope award presents a remarkable opportunity for me work with Dr. Philippe Lefrançois to further enhance my analytical skills for genomics data.”—Parsa Bagherzadeh, HI&DS Award recipient 

“Parsa’s project will identify morphological and dermoscopy features of skin cancers correlated with precision oncology genomics data and clinical outcomes. This will determine visual patterns and molecular biomarkers that can be easily applied to all skin cancer patients, leading to better prediction of disease behavior and patient stratification in a context of limited skin cancer specialist access across Canada.” -- Dr. Philippe Lefrançois, mentor