Adipocytes, or fat cells, serve as reservoirs of energy in the form of lipids in humans, and are tightly regulated with respect to their size and number. Significant alteration in body mass involves alterations in both adipocyte volume and number. Adipocyte-derived factors are significantly increased in obesity and represent good predictors of the development of type 2 diabetes. Moreover, the increase in fat mass has been strongly correlated to the size of the adipocytes, especially in females. In the field of lipid research, the measurement of adipocyte size has served as a good marker for a change in fat mass.

In order for researchers to make discoveries of statistical significance, it is important to have tools that enable them to explore large data repositories in an efficient and flexible manner. Specifically, techniques are needed that can quickly locate adipocytes and accurately compute their sizes, as well as mechanisms that allow researchers to study adipocytes of similar sizes across different data samples. Although there has been related work on automated detection and measurement of adipocytes, they all focused on processing one data sample at a time, and none of them has provided an infrastructure for efficient investigation of a large data collection.

In this paper, we present FatFind, an interactive system that allows the user to search and quantitate adipocytes of different sizes in a large repository of cell microscopy images. The infrastructure that enables this capability is Diamond, a distributed storage system that enables efficient interactive exploration of complex, non-indexed data, detailed in Section 4; the particular techniques for locating and measuring adipocytes are presented in Section 0.

Our approach represents a new method for quantitating adipocytes suspended in a drop under the microscope. Previous efforts to computerize the process have typically involved the use of expensive, proprietary imaging software (such as the Carl Zeiss KS 400) which functions a black box, rather than letting the user control the analysis process. Further, previous approaches have not addressed the need to screen hundreds of thousands of data samples simultaneously, nor do they allow interactive user control. In contrast, both the FatFind application and the Diamond platform on which it runs are open source software that specifically address the previously mentioned needs.