GSoC Developer Adds Direct PACS Server Integration to InVesalius Medical Imaging Tool

·3 views

A Google Summer of Code contributor spent three months building PACS integration for InVesalius, a Brazilian open-source medical imaging platform. Previously, the software lacked native support for connecting to hospital DICOM servers, requiring users to rely on external tools for data exchange. The contributor used Python libraries pynetdicom and pydicom to implement network-based querying, retrieval, and storage of medical images directly within the application. A new user interface was also developed, offering search, multi-level download options, duplicate detection, and graceful error handling with automatic fallback mechanisms. Key technical challenges included managing download cancellation, preventing UI freezes during network queries, and reliably switching between DICOM retrieval methods when needed.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Why HTTP 200 Status Codes Can Hide LLM Policy Failures in AI Gateways

When an LLM gateway's policy layer blocks or flags a message, it may still return an HTTP 200 status code, embedding the refusal inside the streamed response rather than signaling an error — making standard HTTP-level monitoring blind to the failure. This design choice is intentional on chat paths, where a 400 error would show a generic alert and hide the explanation, while a 200 keeps the reason visible to the user in context. The problem is compounded because a single POST request to a chat endpoint can contain an entire conversation history — meaning one HTTP log entry may represent dozens of individual messages, each subject to separate policy decisions. Traditional access logs record one status and one latency per request, but cannot reveal which specific message was blocked, which rule triggered, or whether a policy change caused widespread refusals. Effective monitoring of LLM traffic therefore requires message-level audit records that track policy outcomes per turn, not just transport-level HTTP responses.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

EU Parliament Calls for Stronger Platform Accountability on Cyberbullying

The European Parliament adopted a resolution on 30 April 2026 urging the European Commission to assess whether cyberbullying definitions should be harmonised across member states and whether cross-border criminal provisions are needed. The move is a political signal rather than binding law, and it does not immediately alter legal obligations for businesses or create a new EU-wide criminal offence. The resolution argues that existing frameworks, including the Digital Services Act, may not fully address cyberbullying in all contexts, and calls for more effective platform responsibility. It follows the European Commission's February 2026 action plan, 'Safer Online, Stronger Together', which outlined measures covering education, reporting mechanisms, victim support, and enforcement. Businesses operating online communities or communication channels are advised to monitor the evolving policy process closely rather than assume new compliance requirements are already in force.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Anthropic's Jacobian Lens Tests Whether AI Signals Actually Drive Model Outputs

Researchers at Anthropic have developed a tool called the Jacobian lens to address a long-standing gap in AI interpretability: the difference between a signal being detectable inside a model and that signal actually influencing the model's output. Unlike earlier methods such as the logit lens or tuned lens, which reveal what a layer's state resembles, the Jacobian lens measures whether internal representations do causal work by computing the model's averaged Jacobian from each layer to the final output. The tool includes four operations — READ, WRITE, PATCH, and ABLATE — that together allow researchers to edit internal quantities and observe selective behavioral changes, such as swapping a concept pattern to shift a model's arithmetic answer. Ablating these so-called J-space patterns caused multi-step reasoning to collapse to near zero, though the model retained fluency, sentiment classification, and fact retrieval, indicating the effect is task-specific rather than general. The work highlights that interpretability claims require causal intervention, not just correlation, to be considered reliable.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

LendingChain Launches Zero-Interest Solana Micro-Lending Platform for Emerging Economies

A developer has built LendingChain, a decentralized micro-lending platform on Solana Devnet that offers zero-interest loans ranging from $5 to $25, submitted as part of a DEV Community Weekend Challenge themed around generosity. The platform targets grassroots entrepreneurs in emerging economies, with a focus on clean energy, sustainable agriculture, indigenous women-led crafts, and tech education. Unlike commercial microfinance institutions that charge 25–45% annual interest, LendingChain routes capital directly into non-custodial Solana escrows, keeping costs near zero through Solana's sub-cent transaction fees. Repaid loans are returned to lenders for re-lending in a cycle the project calls the 'Generosity Multiplier,' aiming to maximize the impact of each dollar contributed. The platform is built using Astro, React, Firebase Firestore for real-time state, and Privy for embedded wallet authentication, and is currently live on Firebase Hosting.

0 comments Read more at DEV Community