Simona Turco

Eindhoven University of Technology

Integrating genomics and imaging for improved prostate cancer care

One in seven men is diagnosed with prostate cancer in his lifetime, but only a fraction of these tumors is aggressive and require prompt intervention. Timely and accurate diagnosis is crucial for effective treatment of aggressive prostate cancer and monitoring of low-risk prostate cancer, but this is hampered by the inadequacy of the available diagnostics based on invasive systematic biopsies. With the Hanarth fonds fellowship, Simona Turco will implement a machine learning framework integrating advanced imaging and cancer genomics to achieve non-invasive, accurate diagnosis and prognosis of prostate cancer, and ultimately improve patient outcome by personalized treatment.

Brief summary of final results fellowship 

Although multiparametric MRI (mpMRI) is currently the recommended imaging modality for prostate cancer detection, it is still not accurate enough to replace repeated systematic biopsies. In this fellowship, I aim at harnessing the power of artificial intelligence to improve prostate cancer diagnosis and management. Since the prostate tissue exhibit different characteristics in different areas (prostate zones), it would be useful to segment the prostate zones in the mpMRI. However, this is time consuming and requires manual input from experienced radiologists. To automate this labor-intensive process, we have developed a neural network for zonal prostate segmentation, reaching pixel-wise accuracy up to 98%. In parallel, we are also investigating different methods to extract quantitative imaging features by analyzing spatial and temporal characteristics of the mpMRI sequences. Combining these features via machine learning, we achieved an accuracy for prostate cancerlocalization of 86%. Currently, we are working towards analyzing the genomic makeup of prostate tumors and investigating their correlation with the extracted imaging features. The end goal is to achieve in-vivo, non-invasive tumor profiling by imaging alone, eliminating the need for invasive biopsies.