Bittensor Subnet 76:Safescan

Safe Scan tackles two big problems in AI-driven cancer detection: costly, siloed medical AI and the gap between research and real-world use​. Traditional AI diagnostic tools are often proprietary, expensive, and slow to improve, as med-tech companies focus more on profits and face heavy regulations​. Likewise, many top-performing cancer detection algorithms stay in academic papers or limited grants and never become accessible apps​. Safe Scan’s solution is an open, science-focused approach: it started with some of the best open-source cancer detection algorithms and invites global scientists to continuously improve them​.

To do this, Safe Scan uses Bittensor’s decentralized network as a backbone. Miners in the Bittensor network run Safe Scan’s AI models (for example, image classifiers that detect skin cancer), contributing their computing power. These miners are rewarded in TAO tokens for providing computation​. At the same time, researchers and developers are incentivized to contribute better algorithms – Safe Scan regularly hosts AI model competitions, and the creators of the top-performing cancer detection models earn token rewards​. In essence, Bittensor provides an incentive-based competition marketplace for this purpose, where the “commodity” is accurate cancer diagnoses​. The best model each round earns the lion’s share of token rewards, but with mechanisms to prevent one model from monopolizing (after 30 days at #1, a model’s reward share gradually decreases)​.

End-users of Safe Scan (e.g. patients or doctors using the SkinScan app) benefit from these community-driven improvements at no cost. Users can snap a photo of a skin lesion and have the AI assess cancer risk – all powered by the models running on the distributed miners. Uniquely, Safe Scan also crowdsources data: users who opt to share their dermoscopic images with a doctor’s diagnosis can earn rewards and help expand the public dataset, which further improves the algorithms​. This creates a positive feedback loop: more data and models improve accuracy, which attracts more users and researchers, accelerating progress in early cancer detection.

Technical Architecture and Bittensor Ecosystem Integration

Safe Scan is implemented as Subnet 76 on the Bittensor network, meaning it operates a specialized instance of Bittensor tailored to cancer detection. In Bittensor’s design, each subnet is an independent, incentive-driven AI marketplace that produces a unique digital commodity​. For Safe Scan, that commodity is accurate cancer diagnosis – specifically, AI models that can detect cancers from medical data. The subnet architecture comprises decentralized miners and validators identified by unique network UIDs. Miners run the AI inference tasks (e.g. analyzing images for signs of cancer), while validators test and evaluate the miners’ outputs to determine which models perform best. This whole process is secured and coordinated by the Bittensor blockchain (built on Substrate), which tracks contributions and handles token incentives.

Using Bittensor provides substantial technical advantages for Safe Scan. First, it offers a distributed computing layer: instead of needing a centralized server farm, Safe Scan taps into a global network of Bittensor miners for AI processing power​. This decentralized compute is scalable and cost-efficient, since miners are voluntarily contributing resources in exchange for TAO rewards​. There is no single point of failure or control – the network is robust and censorship-resistant, which is important for an open medical AI service. Second, Bittensor’s built-in consensus and reward mechanism (the Yuma consensus) allows Safe Scan to automatically rank models by performance. Validators in Subnet 76 can challenge miners with known test cases or cross-compare model outputs, and then Bittensor’s consensus logic assigns higher stake (and token rewards) to better-performing models. This creates a self-improving system: the competition framework is essentially baked into the network’s protocol​. Safe Scan’s codebase includes a public leaderboard and evaluation criteria for models, ensuring transparency in how the best algorithms are chosen​.

Architecturally, Safe Scan extends the Bittensor ecosystem into the medical domain. It connects with Bittensor’s tokenomics such that its subnet has a certain allocation of TAO emissions (the dashboard shows Subnet 76’s token metrics, like the “Alpha” token supply and TAO market cap)​. The Safe Scan team configured the subnet parameters (netuid 76) and deployed their custom validation logic (for cancer data) using Bittensor’s Subnet Template as a starting point​. The owner hotkeys and coldkeys of the subnet (visible on-chain) belong to the Safe Scan organization​, which means they manage the subnet registration and can update its settings or add new data for validation as needed. Overall, Safe Scan fits into Bittensor as a novel use-case subnet – one focused on healthcare AI – demonstrating how the Bittensor network can host a variety of AI services (other subnets target language models, image generation, etc., while Safe Scan targets cancer diagnostics).