CRD’s Internship Impact: From Berkeley Lab Cubicles to Television

Last month, Computational Research Division's (CRD's) Daniela​ ​Ushizima​ was an invited guest on the Brazilian television program “Questao de Ordem” where she talked about her work on the analysis of microscopic images. As part of her investigations within the UC Berkeley Institute for Data Science (UC BIDS) she has targeted biomedical imaging such as those provided by the Cell Recognition for the Inspection of the Cervix (CRIC) project. CRIC aims to create computer programs that find abnormal cells in Pap tests faster, more precisely and at a low cost. Three former Lawrence Berkeley National Laboratory (Berkeley Lab) interns are also collaborators on this project: Federal University of Ouro Preto professor Andrea Bianchi (2013), and Federal University of Ceara (UFC) professors Flavio Araujo (2016) and Romuere Silva (2016).  

Sponsored by the Brazilian research program Science Without Borders, the CRIC project—led by UFC professor Fatima Medeiros—brings together an international team of pathologists, cytologists, doctors, professors and scientists to come up with creative solutions—from cervical cancer awareness to Pap smear cell detection algorithms—to mitigate the number of deaths from cervical cancers in the developing world.

As a CRIC collaborator, Ushizima has coordinated software architects and helped facilitate connections between Medeiros and her UFC students with UC BIDS and the Center for Advanced Mathematics for Energy Research Applications (CAMERA) at Berkeley Lab. Medeiros and her students traveled to Berkeley via Brazil’s Benefits program, which provides funding for researchers to collaborate internationally. BIDS hosted the UFC group when they arrived in Berkeley, but the team also explored connections with CAMERA researchers for machine learning algorithms that would not only benefit work on biomedical imaging, but also a broader range of data science applications, such as those at the Advanced Light Source and Molecular Foundry.

Through these Berkeley collaborations, the CRIC team achieved several milestones. They developed:

• SPVD, an awarded classification algorithm for cervical cell image segmentation
• SPVD+, an awarded and improved classification algorithm for cervical cell image segmentation
•  pyCBIR, a content-based image retrieval that works in several domains, including cervical cells
•  Generic algorithms tailored to cervical cells
• The CRIC database, the largest dataset of real cervical images with diverse ground-truth information

“This project is the result of cooperation among researchers who believe that collaborative science can bridge the digital divide,” says Ushizima. “Our team dreams to have automated cervical cell analysis in the public healthcare system and the current strides point out that we are closer than we were when we started years ago.”