1. Radiomics
Non-invasive assessment of tumor phenotype
Precision medicine’s success will heavily rely on integrating various data sources of a patient.
In this context, genomic characterization of individuals has received broad attention as revelation
of the genetic background is hypothesized to uncover molecular events that require adaptation
of therapeutic strategies. In clinical oncology, a number of examples exist where genomics is
already being translated into clinical applications, but these are far behind their expectations.
Recent research has indicated that this is primarily due to intra-tumor heterogeneity, which
cannot be addressed by single biopsies. Moreover, every biopsy is invasive and a burden for the
patient. A promising novel approach in precision medicine, which complements these limitation,
is given by radiomics.
Radiomics is an emerging field that seeks to translate medical images into mineable data by
applying innovative computational approaches to extract a large number of quantitative features
that objectively describe tumor intensity, shape, size, and texture. As medical images capture the
entire tumor volume, are non-invasive, and routinely in use as standard-of-care, it makes
radiomics an ideal source for individualized diagnostics.
To analyze prognostic and predictive value of radiomics, we have developed a computational
framework to extract radiomic data from various imaging modalities (Figure 1). By integrating
these data with genomic information and clinical records, we have recently shown in lung
(NSCLC) and brain cancer (GBM) that radiomic signatures predict for clinical outcomes such as
overall survival or distant metastasis. We furthermore observed strong associations with gene
expression profiles and molecular pathways involved in cell proliferation, immune response, and
p53, which are widely recognized to be of utmost importance in cancerogenesis. Notably, we
also discovered that radiomic features have the potential to predict for druggable somatic
mutations such as EGFR or KRAS. We have found similar results for semantic imaging features
and volumetrics. We further seek to assess whether radiomics may be extended to other human
diseases such as cardiovascular diseases. This may provide an unprecedented opportunity to
advance clinical decision making without additional acquisition risk and at low costs.
Figure 1: Aerts et al., Nature Communications 2014. Tumors show visible phenotypic difference (a),
which can be quantified using radiomics to be integrated with molecular and clinical data (b).
2. Biography
Patrick Grossmann was trained as a classical bioinformatician within the program of
Bioinformatics & Genome Research at the University of Bielefeld (Germany), from which he
received both his Bachelor and Master of Science degrees. His research interest lies in applying
biostatistical methods and machine learning to biomedical data. He earned first data analysis
experience during his Bachelor’s and Master’s Theses at ETH-Zurich, Basel (Switzerland), in the
groups of Computational Systems Biology and Computational Biology, respectively. In addition,
he served as an intern at Dana-Farber Cancer Institute, Boston (USA), during his Master studies.
For the latter two projects, he was awarded two independent scholarships. As a PhD Candidate
at University Maastricht (Netherlands), he now joined Dana-Farber Cancer Institute and Harvard
Medical School to conduct doctoral research on Radiomics-Genomics (finding connections
between quantitative imaging features of tumors and their underlying biology), Toxicogenomics
(understanding how potentially toxic chemicals influence molecular mechanisms of exposed
cells), and Pharmacogenomics (understanding how drug responses depend on the genomic
profile of an individual). Recently, he was awarded the Otto-Bayer scholarship to advance non-
invasive cancer diagnostics by combining Radiomics with Deep Learning. Throughout his
analyses, he puts a strong focus on integrative approaches and reproducible research, thereby
aiming to contribute to the promotion of precision medicine in cancer.
Ways to reach out!
patrick@jimmy.harvard.edu
https://www.linkedin.com/in/patrickgrossmannbioinformatics
https://twitter.com/grossmannpat
http://www.cibl-harvard.org/patrickgrossmann
June 17th, 2016