From Hard Drives to Self-Service: How ASU and Starfish Storage Unlocked 25 Years of Indigenous Health Data

By Allison Proffitt 

October 15, 2026 | For more than two decades, researchers studying one of the world's healthiest aging populations had to wait weeks or sometimes months just to see the data. Now they can find what they need in minutes, browsing folders instead of filing tickets. 

That transformation earned Arizona State University's Center for Evolution and Medicine and its storage partner, Starfish Storage, an Innovative Practices Award at the 2026 Bio-IT World Conference & Expo. Starfish's chief science officer, Ari Berman, accepted the honor on stage and walked the Bio-IT World community winners' session through a project he was quick to credit to someone else. 

Editor’s Note: The 2027 Bio-IT World Innovative Practices Awards Call for Entries has just been launched.  

"I'm going to present work that I had nothing to do with, but it's going to be pretty great," Berman told the audience, before handing most of the credit to ASU's Suhail Ghafoor, IT manager for the Center for Evolution and Medicine, who could not attend. 

A Rare Population, A Growing Archive 

The data at the center of the project comes from the Tsimane Health and Life History Project, a multi-institutional study led by the University of New Mexico and Arizona State University that has tracked the Tsimane, an indigenous forager-horticulturalist community in Bolivia's Amazonian lowlands, since 2001. The Tsimane have drawn intense scientific interest because they show unusually low rates of cardiovascular disease and slower brain aging than industrialized populations, making them a natural laboratory for research into aging, Alzheimer's, and heart disease. 

Since the early 2000s, a hospital in Trinidad, Bolivia, has performed whole-body CT scans on study participants, alongside the collection of demographic, genetic, and biochemical data. Each scan generates hundreds of DICOM files, the standard medical-imaging format, packed with metadata describing scanner settings, anatomy, and protocol. 

The Old Way: Hard Drives And Hope 

Getting that imaging data to researchers used to be a physical, manual ordeal. Scans were copied onto hard drives in Bolivia, mailed to the United States, and loaded onto a central system, where Ghafoor's team would hand-sort files on demand to fill each request. 

"A researcher said, 'I want this data set from these [patients],' and a person had to go in and manually assemble the data," Berman explained. A request as specific as "all high-quality lung scans, males over 50" could take days to fulfill, and with limited IT capacity and a growing list of global collaborators, the queue stretched from weeks to months. 

Complicating matters, 25 years of evolving scanner technology and shifting technician conventions had left the metadata a mess. Body-part descriptions appeared in multiple fields, sometimes in different languages; image quality and scan type were inconsistently recorded; and there was no central catalog tying it all together. 

Layered on top of the technical mess was a governance problem. Because the scans are of human subjects from an indigenous community, every file carried strict consent requirements. Access had to be fully auditable. Because of data-sovereignty and compliance requirements, the system couldn't live in the cloud, it had to run on infrastructure ASU controlled directly, while still supporting both pseudonymized data (re-linkable for longitudinal tracking) and fully anonymized data for broader sharing. 

Building A Self-Service Platform 

Ghafoor, who joined the Tsimane project in 2022, set out to replace the manual pipeline with something researchers could use on their own. As Berman described it, the goal was "a self-service searchable system so they could find exactly what they need without the need to wait in a queue for someone to figure it out" built around directory structures researchers already knew how to browse, organized by body part, scan quality, and reconstruction type. 

Starfish's platform, which catalogs unstructured file data across storage systems at scale provided the underlying engine. It can extract file-level metadata from dozens of file types, apply custom tags, and generate lightweight "virtual" directory views without duplicating any data, all while running entirely on ASU's own hardware to satisfy the compliance requirements. 

The team first automated the data pipeline itself, replacing the hard-drive shuttle with a scripted workflow built on a PostgreSQL job queue that queries, downloads, validates, extracts metadata from, and compresses new scans before transferring them to ASU. "That pipeline runs automatically, replacing months of physical logistics," Berman said, adding that the scripts have since been open-sourced on GitHub for other institutions facing the same problem. 

The harder task was making two decades of inconsistent metadata "speak the same language." Working with consulting radiologists, Ghafoor developed a classification method that identifies which organ or body part a scan captured — bone, lung, heart, brain, body/soft tissue, or neck soft tissue — by matching each file's window center and window width values, numeric imaging parameters, against a clinically validated lookup table. The approach sidesteps unreliable free-text fields entirely, so it works regardless of which scanner or technician produced the file. A Python script runs that classification automatically, tags each file with standardized, queryable attributes in Starfish, and keeps the catalog current as new scans arrive from the field. 

On top of that normalized metadata, the platform generates two views: an internal quality-assurance view organized by anonymized patient, and a researcher-facing view organized by body part, quality, and reconstruction type. Both point to the same underlying files with no copying involved. Access to each view is governed automatically by each researcher's consent tier, so a collaborator only ever sees the datasets their agreement permits. 

Minutes, Not Months 

The results, in Berman's telling, were stark: what once took "months of waiting" became "minutes of self-service fulfillment," with researchers browsing and analyzing data independently rather than filing requests and waiting for staff to act on them. 

According to the award entry, dozens of global collaborators across multiple institutions now discover and access imaging subsets on their own, with request turnaround collapsing from weeks to minutes and manual curation eliminated outright. Ghafoor's team, freed from fielding one-off requests, has been able to redirect its time toward infrastructure improvements, a shift that scales as the collaborator network grows, since each new researcher adds no incremental curation burden. The project also required no new hardware; the main investment was the one-time work of mapping metadata with radiologists, a step that now benefits every subsequent user of the archive. 

"The collaboration between ASU and Starfish transformed a fragmented DICOM repository into a structured metadata-driven resource tailored for research needs," Ghafoor said, in a statement Berman read to the audience. 

A Model Built To Be Copied 

Both the award submission and Berman's talk emphasized replicability. The organ-classification script and its lookup table are posted publicly on GitHub, letting any institution with a DICOM archive apply the same window center/width classification approach independently of the Starfish platform. Starfish's contribution — metadata normalization and virtual directory views — layers on top of existing storage without requiring institutions to migrate data or change formats. 

The team, which also included members of ASU's Center for Evolution and Medicine IT group, Starfish's Matt Hutchinson, and Chapman University's Daniel Cummings, who served as the CT scanning system architect, framed the work as a template for other longitudinal, consent-governed research archives—medical imaging collections, biobanks, and multi-site clinical studies among them—that face the same combination of messy legacy metadata, expanding collaborator networks, and strict access-control requirements. 

For a project born out of two decades of mailed hard drives and manual spreadsheet sleuthing, the payoff is a system that, as Berman put it, gives researchers "a whole lot more access" to study one of the few remaining populations living a pre-industrial lifestyle without anyone on ASU's staff having to hand-carry the data to them.