# RadMesh AI Mode: radiology worklist and AI report drafts Your shift. One conversation. Resident prepares the work. The radiologist decides and signs. A support tool for professionals. Not a medical device. Clinical AI features run in research mode. ## An agentic RIS. Built on AI from the codebase up. It is not a traditional RIS with an AI button. The worklist, the draft, the review and the notices are agents by design. Each one keeps a log; none of them signs. ## A shift in six scenes ### Start Know where to begin. The start screen shows the next study, how long it has been waiting and the queue status, so the shift begins with a concrete action. Priority should have a visible reason, not an unexplained AI score. ### List Urgency has context. The worklist queries Orthanc and distinguishes technologist critical flags, order priority and arrival order. A critical flag deserves prompt attention; it does not prove that the image is abnormal. ### Case Open the case. Keep its context. The case shows the indication, the technologist note and the priors the order declares versus those actually in the PACS. From there it opens on the reading station. Separating what the order declares from what is available prevents a missing prior from being presented as reviewed. ### Voice Dictate without leaving the study. The phone records, RadMesh transcribes it in a couple of seconds and the text enters the chat as an instruction to change the draft. Voice speeds up editing but never signs: signing is always an explicit gesture by the radiologist. ### RAD·QC 12 mm here. 8 mm there. Verify. Every draft version goes through RAD·QC before signature. Measurements and laterality are checked by rules; the model looks for omitted findings, unanswered questions and studies flagged as critical by the technologist whose report calls for a second look. An alert requests review. The radiologist applies the fix or dismisses it with a reason; AI never signs on its own. ### Close Nothing lost between shifts. At the end, the list collects what is left to do: series to repeat, missing priors and second looks. Nothing is sent to anyone today: it is the radiologist's list. One person's shift ends; unfinished work needs to carry its context forward. ## What the clinician reads is the conclusion. A finding that never reaches the conclusion. A clinical question left unanswered. RadMesh looks for them before signing; the radiologist decides. ## Four agents. One shift. Every task in the shift has an agent. Each one logs what it did. None of them signs, and none lowers a priority set by a human. ### Prioritizes The worklist is ordered by turnaround target and by the technologist's flag, with the reason visible on every row. ### Reads the image Every finding, linked to its sentence. The AI marks the area of the image that supports each finding and links it to the sentence in the draft. Hover a sentence and see where it looked. Nothing enters the report without your approval. Research mode. CT today; other modalities in preparation. Each institution chooses which models it uses. ### Reviews Review before signing, in seconds. It looks for findings left out of the conclusion, the unanswered clinical question, laterality and measurements. It shows you the observation and you decide. ### Backup If a study is about to miss its target, it notifies the radiologist on shift. If it is not read, it notifies the administrator to reassign it and suggests who can read. ## For the radiologist - AI Mode by chat or by voice, in Spanish or in English. - Guided reading: "1 of 15"; you move forward, nothing happens on its own. - Draft on your own template (research mode). - Dictation with your template and the procedure code of the order. - Sign in two taps; voice never signs. - Reading mode without distractions. - If the AI does not respond, you keep working. ## Turnaround under control, with a RIS that is yours. - No PACS change: it connects over DICOM from a VM in your network. - Worklist with the reason in view: by turnaround target and by the priority of the order and the technologist. - Reading targets and backup: it warns before a target is missed and suggests who can read. - Connected shifts: it knows who is on shift from Clicrad. - WhatsApp and Slack notices: only to staff on shift, with no patient data. - Service quality: what gets corrected before signing, for training. - Service templates: personal and shared, without junk templates. ## The studies stay in Colombia. Your studies live on a node dedicated to your institution, on servers in Colombia: we host it or, if your hospital prefers, it runs on its own servers. - The RadMesh central service only handles licenses, updates and signed models. It does not store patient data. - Each node is independent: one institution's data never mixes with another's or passes through the central service. - Access with a personal user and mandatory second factor. - Notices never include patient data. - Your studies are not used to train models without a signed agreement with your institution. - Any research use is approved by your own ethics committee. ## How we start 1. Live demo. A 30-minute video call with synthetic cases: the whole shift, from the worklist to the signature. 2. Connection. Your IT team adds a DICOM destination. Nothing else changes in your PACS. We host and operate the node. 3. Pilot with your team. We set up templates, shifts and reading targets, train the radiologists and stay with them during the pilot. ## What RadMesh is, and what it is not. - It is not a medical device; the radiologist decides and signs. - Clinical AI runs in research mode. - Internal evaluations with synthetic cases, not clinical validation. ## Frequently asked questions Do we have to change the PACS or the equipment? No. RadMesh receives the studies over DICOM from your current PACS. Does the AI replace the radiologist? No. It prepares a draft and reviews the wording before signing. The radiologist decides, corrects and signs; nothing signs itself. Where are the studies kept? Your studies live on a node dedicated to your institution, on servers in Colombia: we host it or, if your hospital prefers, it runs on its own servers. What happens if the AI fails or does not respond? The RIS keeps working. RadMesh shows that the review is unavailable and the radiologist reads and signs as always. Does it work in Spanish? Yes. Chat, voice, draft and templates are in Spanish; it also works in English. Who is responsible for the report? The radiologist who signs it, as always. RadMesh is the tool; the clinical decision belongs to the professional. How is it contracted? After the demo we talk with you about how RadMesh fits your service. ## See a full shift in 30 minutes. From the worklist to the signature, with synthetic cases and your own team on the call. https://www.radicalmesh.com/en/ai-mode