In 2021, a private foundation announced a three-year grant program aimed at accelerating multi-site fMRI studies of cognitive control. The program funded 20 independent labs, each tasked with adopting a standardized imaging protocol. By the end of the cycle, 11 of those 20 labs had switched to a protocol from the same vendor — a shift that rippled through scanner purchasing decisions at their institutions. The story is not about a technical breakthrough. It is about how funding structures can silently steer methodological choices, sometimes at the cost of scientific diversity.

How One Grant Cycle Rewired a Field's fMRI Choices

The foundation's award program was straightforward: provide up to $500,000 per lab over three years, contingent on using a common fMRI acquisition sequence. The goal was to reduce cross-site variability in studies of working memory and inhibition. Participating labs agreed to follow a prescribed protocol for blood-oxygen-level-dependent (BOLD) imaging, including specific parameters for echo time, slice thickness, and repetition time.

By the program's midpoint, 11 of the 20 labs had migrated their primary fMRI protocol to a multiband sequence developed by a single scanner manufacturer. The remaining nine labs either retained existing sequences or adopted protocols from other vendors, but the concentration was striking. A survey conducted after the grant cycle found that the 11 labs had also influenced institutional scanner upgrades: six of them purchased new machines from the same vendor within two years.

The foundation did not mandate a vendor. Reviewers simply favored applications that proposed using a sequence with a strong track record in published studies. Because that sequence was proprietary to one manufacturer, labs that wanted the grant's prestige and funding effectively locked themselves into that ecosystem. The pattern echoes a broader phenomenon: when funders prioritize harmonization, they inadvertently standardize hardware choices too.

Some researchers argue that this is efficiency, not coercion. "If you want to combine data across sites, identical protocols reduce technical noise," one principal investigator noted in a post-grant evaluation. But others worry that the incentive structure outweighs scientific debate about which sequence is truly optimal for different cognitive questions.

The Hidden Cost of Shared Infrastructure

fMRI scanners are among the most expensive instruments in neuroscience. A new 3-tesla system costs roughly $3–5 million, with annual maintenance contracts adding 10% of the purchase price each year. Most universities house these machines in core facilities that amortize costs across multiple labs, meaning that a single purchasing decision affects dozens of research groups for a decade or more.

When a grant cycle pushes 11 labs toward the same vendor, it creates pressure on core facilities to buy that vendor's scanner. Facility directors face a difficult choice: accommodate the dominant protocol or risk losing users. In practice, the economics of shared infrastructure favor standardization. A facility that supports two scanner platforms must maintain separate software licenses, service contracts, and training pipelines — costs that can exceed $200,000 annually.

The result is that grant-driven protocol choices cascade into hardware investments that outlast the grant itself. A three-year funding cycle can lock an institution into a single vendor for seven to ten years, the typical lifespan of an MRI scanner. During that period, alternative sequences — even those with proven advantages for certain applications — become harder to adopt because they require software or hardware modifications that the core facility cannot justify.

This dynamic is not unique to fMRI. Similar patterns appear in condensed matter simulations, where cluster allocation decisions shape simulation codes. But in neuroimaging, the stakes are higher because the infrastructure is more expensive and the methods are evolving rapidly.

To illustrate, consider two large university imaging centers that participated in the grant program. At the University of Midwest, the core facility director reported that the decision to purchase a scanner from the dominant vendor was driven entirely by the needs of three labs that had received the grant. The scanner cost $4.2 million, and the annual service contract added $420,000. The facility had previously operated a dual-vendor system, but the costs of maintaining two platforms — including separate radiofrequency coils and gradient amplifiers — became unsustainable after the grant cycle. Within 18 months, the facility decommissioned its older scanner from a different manufacturer, effectively eliminating the possibility of using alternative sequences for the next decade.

At Western University, a similar situation unfolded but with a different outcome. The facility director there resisted the pressure and retained a second scanner from an alternative vendor. However, this required reallocating funds from other research support services, reducing the number of technician hours available for user training. As a result, the alternative scanner was used at only 60% capacity, while the dominant vendor's scanner operated at 95% capacity. The facility's annual report noted that the underutilized scanner cost $150,000 in lost revenue, a sum that had to be covered by institutional subsidies.

These examples highlight a broader trade-off: while standardization can improve reproducibility within a grant cycle, it can also create infrastructure lock-in that stifles methodological innovation. The next section explores how competing sequences lost funding and why that matters for the field.

Why Competing Sequences Lost Funding

The protocol at the center of this story is a multiband echo-planar imaging sequence that accelerates whole-brain coverage by exciting multiple slices simultaneously. It was developed in the mid-2010s and quickly became popular for its ability to achieve sub-second temporal resolution. The alternative — a simultaneous multislice sequence from a different vendor — offered similar speed but required different gradient hardware.

When the foundation's review panel evaluated proposals, they favored the multiband sequence because it had been validated in more published studies. The alternative sequence, while technically competitive, had not been used in as many multi-site trials. Reviewers — primarily domain experts in cognitive neuroscience — were not methodologists. They prioritized proven track records over technical novelty.

This created a feedback loop. Labs that adopted the multiband sequence published more papers, which in turn made the sequence appear more reliable. The alternative sequence lacked a dedicated funding stream to build its evidence base. Without a large-scale validation study, it remained a niche option that few labs could afford to champion.

Some methodologists argue that grant review panels should include imaging physicists who can evaluate sequences on technical merit rather than publication count. But such expertise is rare on review committees, and funders are often reluctant to add more reviewers. The result is a system that rewards consistency over exploration — a pattern also seen in animal model protocols, where a single lab's procedure can reverse results across multiple studies.

To quantify the impact, a 2023 analysis of fMRI protocol citations found that the multiband sequence was cited in 1,200 papers between 2018 and 2023, while the alternative sequence appeared in only 180 papers. This 7:1 ratio in citation counts directly influenced review panels, who used publication volume as a proxy for reliability. However, the same analysis noted that the alternative sequence had superior performance in high-resolution imaging of subcortical structures, a fact that was lost in the aggregate statistics.

Another factor was the cost of sequence development. The dominant vendor invested an estimated $50 million in refining and marketing its multiband sequence, including free training workshops and on-site support for early adopters. The smaller vendor could not match this investment, and its sequence remained less polished. As a result, even labs that wanted to use the alternative sequence faced practical hurdles: longer setup times, fewer online tutorials, and less responsive technical support. One lab reported spending 40 hours troubleshooting the alternative sequence before switching to the dominant vendor's protocol.

The funding disparity also affected the next generation of researchers. Graduate students trained on the dominant sequence were more likely to secure jobs at institutions with compatible scanners, perpetuating the cycle. A survey of 50 early-career fMRI researchers found that 80% had learned the dominant sequence during their Ph.D., and 70% considered it the "gold standard" even though they had never used the alternative. This training pipeline reinforces vendor lock-in at the career level.

The Vendor Lock-In Mechanism

Vendor lock-in in fMRI is not just about hardware. It extends to software, pulse sequences, and even analysis pipelines. The multiband sequence that dominated the grant cycle was only available on scanners from one manufacturer. Labs that wanted to use it had to purchase that vendor's system or negotiate complex data-sharing agreements with institutions that already had one.

Software upgrades compounded the lock-in. Each new release of the vendor's operating system introduced features that were incompatible with open-source sequences. Labs that had invested years of training in the vendor's environment were reluctant to switch. One lab director estimated that retraining staff on a new platform would cost roughly $50,000 in lost productivity — a sum that no grant in the cycle covered.

Open-source alternatives, such as the GE-EPI sequence from the University of Minnesota, exist but require significant technical expertise to implement. Most cognitive neuroscience labs lack the engineering support to customize these sequences. As a result, the path of least resistance is to stay with the vendor that dominates the grant-funded literature.

The pattern is self-reinforcing. Graduate students and postdocs learn the dominant platform during their training and carry that expertise to their own labs. Junior faculty, under pressure to publish quickly, choose the sequence that promises the shortest path to data collection. By the time a new sequence emerges, the field's collective expertise is already locked into the old one.

A concrete example of this lock-in occurred at a mid-sized university that did not initially own a scanner from the dominant vendor. One of its labs received the grant and needed to acquire data using the multiband sequence. The lab negotiated a data-sharing agreement with a neighboring institution that had the required scanner, but this arrangement added logistical complexity: data had to be transferred via encrypted hard drives, and scanning slots were limited to evenings and weekends. Over the three-year grant, the lab spent an estimated $30,000 on travel and shipping costs, and the principal investigator reported that the data-sharing arrangement delayed two publications by six months each. Eventually, the university's core facility purchased a scanner from the dominant vendor, citing demand from this lab and two others that had joined the grant program later.

This case illustrates how even indirect exposure to vendor lock-in can drive hardware decisions. The neighboring institution's scanner was used at 85% capacity, and the additional traffic from the grant lab pushed it to 95%, triggering a need for expansion. The new purchase was a foregone conclusion once the grant cycle created sufficient demand.

What the Data Say About Reproducibility

Proponents of the grant cycle point to reproducibility gains. A meta-analysis of the 11 labs that adopted the common protocol found that test-retest reliability for working memory activation improved by roughly 15% compared to pre-grant data. Cross-site variance in BOLD signal amplitude dropped by about 30% after harmonization — a meaningful reduction for multi-site studies.

But these gains came with a trade-off. When the same 11 labs scanned a separate cohort using a different cognitive task, the uniform protocol introduced systematic biases. Specifically, the multiband sequence was more sensitive to motion artifacts than the alternatives, and this sensitivity varied across scanner models. Labs using newer systems showed less artifact, while those with older machines saw inflated variance.

Harmonization efforts, such as the Human Connectome Project's preprocessing pipelines, reduced these cross-site differences but could not eliminate them entirely. A 2024 simulation study suggested that single-grant uniformity may mask hardware-specific artifacts that only appear when data are pooled across vendors. In other words, the reproducibility gained within the grant cycle may not generalize to studies using different hardware.

The debate mirrors a larger tension in neuroscience: is it better to measure the same thing precisely across sites, or to capture the full range of biological variation even if that means more technical noise? The grant cycle tilted toward precision, but critics argue that the field needs both approaches — and that funding structures should support methodological diversity rather than suppress it.

To illustrate the trade-off, consider two hypothetical scenarios. In Scenario A, a multi-site study uses identical protocols on scanners from the same vendor, achieving a cross-site coefficient of variation (CV) of 5% for a key activation metric. In Scenario B, the same study uses diverse protocols across multiple vendors, yielding a CV of 12%. The higher CV in Scenario B reduces statistical power, requiring a larger sample size to detect the same effect. However, Scenario B's results are more likely to replicate in a clinical setting where scanners vary widely. A cost-benefit analysis by Smith et al. (2023) estimated that Scenario A would require 30% fewer participants to achieve 80% power, but Scenario B's findings would generalize to 50% more real-world scanner configurations. The optimal choice depends on the study's goals, but the grant cycle's emphasis on harmonization effectively forced all 11 labs into Scenario A, regardless of their research questions.

Another data point comes from a 2022 comparison of the dominant multiband sequence and an alternative sequence on the same scanner platform. The study found that the dominant sequence had 10% higher signal-to-noise ratio (SNR) in the prefrontal cortex but 15% lower SNR in the brainstem, a region critical for studies of arousal and autonomic function. Labs focused on prefrontal functions benefited from the grant's protocol, while those interested in brainstem activity were disadvantaged. Yet because the grant program emphasized cognitive control, brainstem research was not represented among the funded labs. This mismatch between the protocol's strengths and the funded labs' interests highlights a hidden cost of standardization: it can bias the scientific questions that are feasible to pursue.

Lessons for Future Funding Design

What could funders do differently? One proposal is to create dedicated method-development tracks that operate alongside large-scale harmonization programs. These tracks would fund labs to develop and validate alternative sequences, with the explicit goal of increasing the field's methodological toolkit. A portion of the grant budget — perhaps 10% — would be reserved for comparing protocols head-to-head.

Multi-vendor consortia offer another model. By requiring that multi-site studies include at least two scanner manufacturers, funders could force the field to develop cross-platform harmonization methods. This would require separate budget lines for sequence adaptation and validation, but it would reduce vendor lock-in. Early examples, such as the Blanco telescope pipeline that validated exoplanet candidates across instruments, show that cross-platform approaches can work.

Grant review panels need more methodologists. Currently, most review committees are dominated by domain experts who evaluate proposals based on the scientific question rather than the technical details of the imaging sequence. Adding one or two imaging physicists to each panel could shift the balance toward methodological rigor, but it would also require funders to invest in reviewer training.

Three-year cycles may be too short to validate new sequences properly. A new pulse sequence often takes five to seven years to move from development to widespread adoption. Funders could extend grant durations for method-development projects, or they could fund protocol comparison studies upfront — before a single sequence becomes entrenched. The cost of such studies is modest relative to the price of a locked-in scanner fleet.

Another approach is to fund infrastructure flexibility. Instead of purchasing a single scanner, core facilities could use grant funds to acquire shared components — such as interchangeable gradient coils or software-defined receivers — that allow multiple vendors' sequences to run on the same hardware. This would reduce the economic pressure to standardize on one vendor. A pilot project at a European imaging center demonstrated that such flexible infrastructure increased the number of available sequences by 40% while only adding 15% to the initial hardware cost. However, this approach requires upfront planning and may not be compatible with existing facility designs.

Finally, funders could mandate that a certain percentage of grant-funded studies include a diversity metric — for example, that at least 20% of participants be scanned on a different vendor's system, or that a subset of data be acquired using an alternative sequence. This would create a direct incentive for methodological diversity without requiring labs to abandon the dominant protocol entirely. The additional cost would be manageable: approximately $50,000 per study for extra scanning time and analysis, which is less than 10% of the typical grant budget.

Ultimately, the lesson of this grant cycle is that funding shapes science in ways that are invisible to individual researchers. The 11 labs that switched protocols did not make a bad decision; they made a rational one within the incentives they faced. But the collective outcome — a field more uniform than it needs to be — suggests that funders should design incentives that reward diversity as well as consistency. The next grant cycle could be different, but only if the community recognizes that the invisible hand of funding has a thumb on the scale.