Gabriele Campanella

Gabriele Campanella

Gabriele Campanella

Memorial Sloan Kettering Cancer Center
417 East 68th St., Z-682 area
New York, NY 10065
gac2010@med.cornell.edu

PhD Student

I received my bachelor's and master's degrees at the Universita Politecnica delle Marche (Ancona, Italy) in biology. In 2015 I moved to New York to pursue a Ph.D. at Weill Cornell Medicine and joined the lab of Prof. Thomas Fuchs in November 2016. My research experience before starting my PhD was mostly focused on biophysics, in particular the analysis of small angle scattering experiments. Coming to the lab of Prof. Thomas Fuchs at MSKCC I've transitioned to the field of computer vision, where we use deep learning techniques to advance the way cancer is treated.

Personal CV

Projects:

Quality control for digital pathology

The quality of virtual pathology slides is becoming of more and more importance as deep learning approaches are applied in the pathology domain. Unfortunately, scanning systems fail to provide sufficient quality frequently generating slides with out-of-focus regions. Manual quality control is the current standard to deal with poor data quality. This is time consuming and impractical for the needs of current computational techniques that require large datasets.

To tackle this issue I am developing a quality control pipeline to determine the quality of a virtual slide and to determine whether an image patch extracted from such slides are of sufficient quality for subsequent computational treatment. In a recently submitted manuscript we describe how a random forest approach with 13 engineered features can be effective to determine the quality of an image in terms of sharpness. We also compared this classical machine learning approach with deep learning techniques in which the features important for classification are learned during training.

Basal Cell Carcinoma (BCC) and melanoma

BCC is the most common form of skin cancer in the world, affecting 800,000 people every year in the US alone. While it is not generally particularly difficult to detect and treat, BCC constitutes an ideal domain to develop deep learning models for tumor detection, classification and grading. We are combining the quality and large amount of data at MSKCC with state-of-the-art convolutional neural networks to detect tumor regions in virtual slides. Importantly, the methods developed on this domain can be translated to other types of malignancies, in particular melanoma.

Immunotherapy has proven to be a highly effective treatment for distant stage melanomas. Nevertheless, not all patients benefit from these tehrapies and much research is undergoing to find markers to predict immunotherapy efficacy. Our long-term goal for this project is to guide the treament decision based on morphological features of the malignant tissue.