RadFuzion 7.0.0
23 new29 improved22 fixed6 security & privacy
New
- Modality simulator. Troubleshooting › Modality Simulator acts like a scanner, so RadFuzion (or a PACS through it) can be tried without a real modality: at go-live, for training, or for a demo. Pick a setup (X-ray, CT, MRI, ultrasound, mammography, NM, PET and more; save your own), check its connections, query the modality worklist and pick an order (or type one in), and do the exam: MPPS In Progress, the images sent under the order's patient and accession, MPPS Completed or Discontinued, and storage commitment if asked. Each step shows what the other side answered. The images are made-up test patterns with SIMULATED burned in; pictures (a camera too), DICOM files or a folder on the server can be sent instead, stamped as the exam's. Worklist, images and MPPS can each go to this RadFuzion, a DICOM destination or another system. Looking needs Troubleshooting view; sending needs Transfers manage, and each exam sent is in the event log and the access log.
- Organ outlines in the viewer (TotalSegmentator add-on, free). Settings › Add-ons installs TotalSegmentator, an open-source organ segmentation model, with one click: about 3 GB downloaded once from where each part is published (Python from GitHub, PyTorch for the CPU from pytorch.org, TotalSegmentator and its models from PyPI and GitHub), never part of RadFuzion's own install. Then a CT or MR series' Organs button outlines its structures (over 100 on CT, about 50 on MR: organs, bones, vessels, muscles) in a minute or a few on the server's CPU, at low priority so routing isn't held up, one series at a time. The outlines go on the images with the other results (each structure can be hidden), with each structure's volume, and export as a volumes CSV, NIfTI or a DICOM SEG. Point at a structure, with any tool, and its name ("Right kidney", "L1 vertebra") shows beside the pointer with a short line to it, always on screen however the image is zoomed or panned: a quick way to learn the anatomy. Organs is on each viewport's right-click menu (no trip to the toolbar); once a series is outlined, a small Organs button on its viewport shows or hides the outlines, and the series strip marks the series already outlined. The outlines open on the key organs (brain, lungs, heart, liver, spleen, kidneys, pancreas), with Organs, Skeleton, Vessels, Muscles and All a click away. They're kept with the exam: reopen it, even after a restart, and they show again at once, for as long as its images are on the server (when the viewer cache or the image cache lets the exam go, they go too). On CT, and on an MR outlined with MR spine, right-click › Spine labels from TotalSegmentator puts a spine label (C1 to S1) on each vertebra it found, on the sagittal series. More structures download on request in Settings › Add-ons (about 250 to 300 MB each): MR spine (each vertebra, labeled on MR too), the liver's eight segments on CT and on MR, head and neck bones and vessels, the head's glands and cavities, and the lungs' vessels and airways; the Organs panel then offers each one for the series, and switching back to the organs is instant. Only TotalSegmentator's open (Apache-2.0) anatomy tasks are used (none that look for disease), its usage statistics are off, and nothing runs until someone clicks. Not for diagnosis.
- Each facility's priors go to its own PACS. RadFetch facility overrides (matched on an HL7 field such as MSH-4) can now send the priors they fetch to that facility's own destinations instead of through the routing rules. Facilities that share one archive but are separate places at the PACS each get their priors where they read: one archive searched, each prior sent straight to the facility that asked for it. When two facilities ask for the same prior at once, both get it; a facility that asks later gets its own copy without the first being sent it again. Skip If Already On Destination checks only that facility's own destinations, and a facility whose destinations are deleted or turned off shows in Settings checks. Facilities left on Routing rules work as before.
- See where each Query/Retrieve answer came from. When workstations query RadFuzion's Query/Retrieve AE with archives passed through, each answer now says where the exam is: Instance Availability is ONLINE when it's in RadFuzion's image cache and NEARLINE when only an archive has it (a retrieve pulls it first), which many viewers show as a column. You can also have the archive's name, or the cache's, written into a field the workstation shows (Institution Name, Station Name, Institutional Department Name or Study ID), in Settings › RadCore › Query/Retrieve › Show Each Answer's Source In. Only the query answers change, never the images. Writing the source into a field is off by default.
- Workstation Activity. Query / Retrieve now lists what other systems did on RadFuzion's Query/Retrieve AE: each search with what it asked and how many answers came from the cache and from each archive (and which archive couldn't answer), each retrieve with the exam, where its images came from (the cache, or pulled from an archive), how many reached the workstation and how long it took, and every caller turned away. Filter to problems, or search by AE title, address, patient or archive. It keeps the last 500 until the service restarts and needs the Query/Retrieve view permission; clearing it needs manage.
- Hanging protocols in the viewer. Exams now open by modality instead of one viewport per series in series-number order, so an OT or SC series (scanned paperwork, screen captures, dose sheets) never hangs first or lands among the images: those go to the end of the series strip, after the scouts and localizers. Mammograms hang RCC and LCC over RMLO and LMLO, back to back (2D before synthesized 2D before tomosynthesis); radiographs hang one view per viewport, frontal before lateral, even when the views share a series; CT shows the axial, coronal and sagittal series and the lung or bone kernel; MR shows its first six sequences without localizers or calibration scans; PET/CT shows corrected PET first. Hang in the viewer's top bar re-hangs the exam (Alt+H) or switches back to every series in order. The patient CD viewer is unchanged.
- ECGs in the viewer. A 12-lead ECG, or another DICOM waveform (hemodynamic, electrophysiology, arterial pulse, respiratory), now opens on its own sheet instead of failing as an image: the traces on the grid, laid out 3×4 with a rhythm strip, 3×4, 6×2 or 12×1, at 12.5, 25 or 50 mm/s and 5, 10 or 20 mm/mV, with the cart's measurements (rate, PR, QRS, QT, QTc, axes) and statement above them. An ECG-only exam opens straight onto it.
- Videos in the viewer. Endoscopy, ultrasound and surgical clips stored as MP4 (H.264 or H.265) play in the viewer with frame stepping, 0.25× to 2× speed and loop, and can be measured and saved a frame at a time. MPEG-2 video and transport streams, which browsers can't play, say so and can be downloaded.
- AI results on the images. A segmentation (SEG), an RT structure set or a measurement report (SR) is drawn on the series it was made on, as colored masks, outlines or measurements: click its series, or Show on images on the report. Each result can be hidden, faded or removed, and each segment or structure hidden on its own, from the Annotations panel.
- Undo and redo in the viewer. Ctrl+Z and Ctrl+Y (and the toolbar) undo and redo measurements drawn, moved or deleted, Clear All included, and segmentation.
- Inspect: troubleshooting tools in the viewer. For PACS admins and radiology IT checking whether a display problem is in the images or in the PACS. Inspect in the viewer's top bar opens a panel beside the images:
- Check reads every header in the series, or the whole exam, and says in plain words what breaks stacking, MPR or filing: slices missing (how many, and where), uneven spacing, images tilted differently, gantry tilt, several stacks in one series (echoes, phases), Instance Numbers repeated or out of slice order, one SOP Instance UID on two images, a patient, study or accession that changes within the series or between series, mixed image sizes, encodings and pixel formats, and fewer images than the scanner made. Click an image number to show it, or a tag to find it.
- Compare lists the tags that differ between two images, or across the whole series, with the images that have each value; tags that normally change per image are set aside. Two images' pixels are compared too: identical (even stored differently), or how many values differ and by how much, with a map of where. Compressed images are decoded by a separate worker process, so a damaged one can never stop the server. Two series compares where they are in the patient, from their headers: their planes, grid, spacing and frame of reference side by side, the angle between them, how far apart their centers are and how much they overlap, and whether they line up as they are or only registration can line them up.
- Validate checks files against the rules strict PACS enforce: File Meta that disagrees with the dataset, the tags the DICOM standard requires of that kind of image (every mandatory module of its IOD, for each of the standard's 164 storage SOP Classes: a mammogram's Positioner Type, a CT's KVP) missing or empty, values that break their VR (dates, times, UIDs, code strings, lengths), the wrong number of values, private tags without a creator, and pixel descriptions that can't be right.
- Display shows every header value that decides how the image looks (bits, sign, rescale, each header window, VOI LUT function and tables, MONOCHROME1, presentation state, overlays, shutters) next to what the viewer is doing, and switches one step at a time.
- Changes shows what was changed in the image after it was made (its Original Attributes record: before and now, by which system, when), and RadFuzion's change log for the exam.
- Trail shows where the exam came from, how it was routed, where it went, and its events.
- Export saves the original file, a de-identified copy for a vendor ticket (new UIDs, dates shifted, device details kept, private tags optional; it warns when the pixels may hold names), or the tags as text or JSON. Every export is in the Access Log with the exam and image.
Inspect needs Tag Editor view.
- Check a whole exam from Transfers. Right-click an exam › Check Exam runs Inspect's Check and Validate on every series and image at once, and more: a few images per series are decoded to catch pixel data that's damaged or cut off, and reports, Key Objects, presentation states, segmentations and RT structure sets pointing at images that never arrived are listed. Problems and warnings are per series, in plain words, with the images they're on; the last check of each exam shows on its Transfers row. Download the report as a page (to read, print or attach to a ticket) or CSV, with or without patient details. A routing rule can run it on arrival (Check on Arrival) and notify when it finds problems (notification event "Exam check found problems"); with Hold Exams with Problems, those exams aren't sent until someone reviews the findings and sends them anyway, from Transfers or the Dashboard. Warnings never hold, and a check that can't run never holds an exam. Needs Tag Editor view; sending a held exam needs Transfers manage.
- Compare an exam with an archive's copy. Right-click an exam on Transfers › Compare with Archive pulls the exam back from a Query / Retrieve source (straight from the archive, never from RadFuzion's own image cache) and compares it with RadFuzion's copy image by image: images missing in the archive (or only there), tags that differ (with each value here and there; tags an archive keeps itself, like Retrieve AE Title, left out), pixels that differ, and how each copy is stored. Pixels are compared decoded, so a lossless re-encoding (the archive storing JPEG 2000 Lossless what was sent uncompressed) counts as identical, and a lossy one is named as the likely reason when values differ a little. It runs on the server with progress; the result is kept with the exam and downloads as a page or CSV. The pulled copy is deleted when it's done. Needs Tag Editor view.
- Organs in 3D. Once a CT or MR series is outlined (TotalSegmentator add-on), MPR's 3D pane has an Organs button: every structure as a smooth surface, by group (Key organs, Organs, Skeleton, Vessels, Muscles, All), with or without the volume rendering and see-through on request. Point at a structure for its name; click it to move the three planes there and open a short card: where it is, what it does and what's next to it, in plain words, for learning (not for diagnosis). The surfaces are made on the server once, in about ten seconds, and also export as STL (every structure zipped) from the Organs panel, for 3D programs and printing.
- Find and black out burned-in text (Pixel PHI add-on, free). Settings › Add-ons installs Pixel PHI (about 250 MB: Python from GitHub, then RapidOCR with PaddleOCR's open models, ONNX Runtime and OpenCV from PyPI). Then Transfers › right-click › Find Burned-In Text reads the text in an exam's images (ultrasound, screen captures, X-ray, scanned pages and anything that says it has text; or every image), on the server's CPU at low priority. Each image is shown with what was found: the patient's name, IDs, the accession and dates from the header in red and chosen to black out, other text (a probe frequency, a side marker) kept; click a box to change it, or drag to add one. Download redacted copy gives the scanned images with the chosen regions black in every frame of a cine, the header de-identified (new UIDs, dates shifted) and marked derived. Send redacted copy sends that same copy to a destination you pick: it shows on Transfers as an exam of its own (named REDACTED- and ten characters), goes straight there without the routing rules, and the original exam is not sent or changed. The copy has no patient name or birth date, so the required-tag check does not hold it. A device template (manufacturer, model, image size) remembers where a machine always prints its text, and is chosen on its images from then on. The originals never change. OCR can miss text, so the review says to check every image. Needs Tag Editor view; sending the copy needs Transfers manage as well; templates need Rules manage.
- Line one series up with another (Image Registration add-on, free). Settings › Add-ons installs Image Registration (about 150 MB: Python from GitHub, then SimpleITK, open source, from PyPI). Then the viewer's Inspect › Compare › Two series › Line up moves another series onto the one showing (a prior onto today's scan, an MR onto a CT, two X-rays) by mutual information, rigid (turn and shift) or affine (also scale and shear), in seconds on the server's CPU, and shows how well: the turn and shift it took, how closely the two match before and after, the two overlaid in color (gray where they agree) before and after, and a checkerboard, through the slice and across it. The moved series downloads as NIfTI. Compressed images are read in the add-on's own process, so a damaged file can't stop the server; results are kept a day, for whoever ran them, and the download is in the Access Log. Needs Tag Editor view. Not for diagnosis.
- Overlays and display shutters in the viewer. Overlay planes stored with an image (CAD marks, modality annotations) are drawn over it, and an X-ray's display shutter (rectangle, circle or polygon) blacks out what's outside it, as a PACS shows them. Both can be turned off in Inspect › Display.
- Search an image's DICOM tags. The viewer's Tags window (now DICOM Tags) has a search box: type a tag number in any usual form (0018,0050, (0018,0050), 00180050, or just a group like 0019), a name or keyword (slice thickness, KVP), or part of a value, and every match is highlighted; Enter and Shift+Enter step through them, and Ctrl+F jumps to the box. It now lists every tag: sequences open into their items (and a match inside one shows where it sits), each tag's VR, File Meta, private tags (marked, and named when known), and empty tags, which used to be left out. Private only and Hide empty narrow the list, and each value can be copied. Users with Tag Editor view can search; editing still needs Tag Editor manage, and only the image's own top-level values can be edited.
- More viewer tools. W/L Box sets the window from a box you draw; Height measures straight up and down; Eraser removes the measurement you click; Lung draws the pleura and B-lines on lung ultrasound and shows how much of the pleura the B-lines cover; Fill fills the tissue you click and everything connected that looks like it, on any tissue, bright or not (drag right or up for more, left or down for less), and labels it with its area, mean, SD and range, in HU on CT (for a quick look, not diagnosis); and in MPR with MIP, Jump moves the other planes to the brightest point you click.
- Processing workers: another machine's GPU for organ outlines. RadFuzion usually runs on a server or a VM with no GPU. Settings › Processing Workers adds a worker: an Apple silicon Mac (macOS 14 or later) or a 64-bit Linux machine, set up by one command pasted into its Terminal, running as a background service with no window. With Settings › Add-ons › TotalSegmentator set to run outlines on a worker, a CT or MR series is outlined there, and on a GPU worker the organs can run at full resolution (1.5 mm in place of 3 mm, a switch, off by default). In our test on a Mac Studio (a 105-slice chest CT), full resolution took under a minute on the Apple GPU against over five on its CPU, with the same result. The worker calls the server over HTTPS and trusts only that server's certificate; nothing is opened on the worker. The series is sent without the patient: each image is rebuilt from its pixels and geometry, and the worker deletes its copy when the job ends. Registration uses a single-use token (an hour), each worker has its own credential, and Remove ends it at once. With no worker online, or one that goes quiet mid-job (a Mac asleep), the outline runs on the server as before, and the Organs panel says where it ran. Images sent are in the Access Log; registrations, jobs and fallbacks in the event log. A Linux worker with an NVIDIA card (64-bit Intel or AMD, the card's driver installed) installs PyTorch's CUDA 12.6 build from its own locked list and runs on the card; any other Linux worker runs on its CPU.
- Route to Existing Study. A routing rule can send an exam to the archives that already have its study. It is for results that come back to one central gateway, such as a post-processing workstation's marked-up images, and belong at whichever facility holds the original. On the rule's Routing tab, pick the archives to look in (DICOM nodes by C-FIND; DICOMweb, Google Cloud and Azure DICOM stores by QIDO-RS) and what to match on: the Study UID, the accession number, or both. Before the exam is sent, each archive is asked; it goes to every one that has the study, beside the rule's own destinations if it has any (a destination in both gets it once), and a rule can have this alone with no fixed destination. When none has it the exam is held and nothing is sent, to the fixed destinations either: it shows on the Dashboard and on Transfers as Held, where Choose destination shows what each archive answered and sends it where a person says (retries and late images follow that choice; Ask Again asks the archives once more). An archive that doesn't answer is never taken to have said no: it is asked three times, then the exam is held the same way, even if another archive has the study. It can't be combined with Nowhere (Drop). A new notification event, "Route to Existing Study: exam held"; Setup Check flags a rule whose archive is off, gone or can't be asked; holds and choices are in the event log.
- Organ outlines by rule. A routing rule has an Organ Outlines switch (last on its Processing tab, offered while the TotalSegmentator add-on is installed and switched on): each exam the rule matches gets one series outlined by itself after it is sent, so the outlines are there when it's opened in the viewer: drawn on the images as the exam opens, with no click. So a rule can't overdo it, the limits are fixed: CT and MR only, axial only, and only the exam's first axial series; never a coronal, sagittal, oblique or mixed-plane series, a localizer, an enhanced (multi-frame) object or a series under 10 images (any series can still be outlined by hand in the viewer). It is the last thing done with an exam: after it is sent, one exam at a time (200 waiting at most), at low priority, behind anyone's own click on Organs. Routing and forwarding never wait for it, and an outline that fails is noted in the log and nothing else. The exam has to stay on the server: the image cache, Archive Forwarded, or a drop rule with Keep in Image Cache. The outlines stay in RadFuzion: they are never forwarded, resent or added to the exam's images, and a DICOM SEG is written only when someone downloads one. Transfers shows where each exam's outlines stand, live: waiting (with its place in line), outlining, Organs ready, failed, or why nothing was outlined. Setup Check says when the add-on is off or exams aren't kept; the Organs panel and the event log name the rule.
- Drop rules can keep what they drop. A rule sent to Nowhere (Drop) has a Keep in Image Cache switch (off by default): the exams it matches go nowhere, but stay in the image cache (and leave it like any cached exam), so they can be opened, checked or resent while troubleshooting. Late images join them; Transfers says the exam was kept, and the rule's card shows Kept in Cache. Needs the image cache on; Setup checks says so when it isn't.
Improved
- Mammography CAD on the images, on a button. Exams with a mammography CAD report have a CAD button (Alt+Shift+C) that draws its marks on every view, including tomosynthesis (on each slice when the report names none). CAD reports often name images that aren't in the exam (Hologic's tomo CAD runs on secondary captures of the synthesized view): their marks now go on the exam's image of the same breast and view, scaled by the detector spacing, where before nothing could be shown. The findings are listed under Annotations › CAD with their views; a click goes to the image and highlights it. Breast density is listed once, and the individual calcifications a report leaves optional show with All marks. On or off is remembered, so once on CAD comes up by itself on the next exams.
- Tomo one click from the 2D view, every view in step. A mammogram view with tomosynthesis has a Tomo button: its slices replace the view and play from the first, and 2D goes back. Tomo on the toolbar (Alt+`) loads every view's slices first (the button shows how far), then plays them all together from slice 1: a view with fewer slices moves a little slower, so all of them reach the ends at the same moment. Tomo again puts every view back to 2D. The slices load in the background as soon as the 2D views are up, so Tomo plays at once instead of loading first. Scrolling a view while they play takes over: they stop, and scrolling one moves the others to the same place in their own slices. While several views play, their slices load at the size they show on screen (a quarter of the memory or less, and much quicker to load) and come back at full size when they stop. The Tomo Unstable warning is gone.
- Mammograms hang edge to edge, zoomed to the breast. Each breast's chest wall now sits against the viewport's edge, so a pair meets in the middle the way mammography workstations hang them (tomosynthesis too, flipped like its 2D view), zoomed so the breast fills the viewport instead of the black around it. The corner text moves to the outer side, off the image; on a narrow screen, where the image fills the viewport, it stays in its corners. To see the whole image: press and hold Zoom for Fill, Whole image or Actual pixels (1:1), press Alt+0 to switch between fill and whole, or right-click. Reset brings the default back, and Zoom's menu has a remembered switch for whether mammograms open filled.
- The mouse no longer switches viewports on its way somewhere. A viewport becomes the active one when the pointer rests on it a moment (a quarter of a second); a click, a press or the wheel makes it active at once. Moving across viewports to the toolbar no longer changes which one the buttons act on.
- This image's panels on the right-click menu. DICOM tags, Info, Inspect, overlay text, Annotations, Hang and the keyboard shortcuts are now at the bottom of the right-click menu, whatever the tool, for the image right-clicked: the mouse no longer crosses other viewports to reach the top bar and switch the image they open on. While Info or Inspect is open, moving the mouse no longer switches the viewport they show; a click does.
- Transfers as a spreadsheet. Transfers' CSV button saves the exams listed (as filtered) as a CSV file, with or without the patient's details, and right-click an exam › Series List (CSV) or Image List (CSV) lists what it holds. Each export is in the Access Log.
- Save what the viewer shows. Right-click a viewport › Save image (PNG or JPEG) saves the view as it looks, measurements included, and Save series as NIfTI saves a stack of slices as a volume for research and AI tools (Tag Editor view; slices missing or unevenly spaced are refused rather than closed up). Both are in the Access Log.
- Cine speeds, and Auto by default. Every exam opens with Cine on Auto, which plays at a speed that suits what is on screen. CT, MR and tomosynthesis go by how many images the series has: a CT at 5, 10, 15, 20, 30, 40 or 50 frames a second (up to 50, 100, 200, 400, 700 and 1,000 images, then more), a routine MR at 5, 8, 12, 15, 20, 25 or 30 for the same sizes, tomosynthesis at 15 (up to 30 images), 25 (up to 80) or 30. Ultrasound clips, cardiac cine MR and angiography or fluoroscopy play at the speed they were recorded at (30, 20 and 15 when they don't say). Press and hold Cine for 10, 20, 30, 50, 100 or 200 frames a second, on the toolbar or under More; the speed picked lasts for the exam on screen. Fast speeds skip images to keep up instead of slowing down, and scrolling the viewport that's playing stops it, on the image scrolled to.
- Hold for the 3D cursor. With any tool picked, hold the left button still for half a second and the mouse is the 3D cursor until you let go; then it is your tool again (as both buttons held are window/level). A drag that starts at once stays the tool's own. Not while magnifying, filling or erasing, on a measurement under the pointer, or in MPR.
- Organs in MPR, whichever series you open. The 3D pane now offers an exam's organ outlines when MPR is opened on a different series than the one outlined (a rule outlines the first axial series), as long as both lie in the same frame of reference. The organ options moved up, clear of the Z Y X clip buttons.
- Reading a CAD report. The report sheet starts a mammography CAD report with what the CAD looked for and marked (clusters, masses, calcification count, density), with the full report folded below it. Deep reports no longer push their values into a narrow column at the right edge where the words stood one per line. CAD reports are named as such in the Reports list before they're opened (from the document's title or its SOP class) instead of Structured Report.
- Keyboard shortcuts that never type into a report. Every viewer shortcut is now an Alt combination (Option on a Mac): Alt+W window/level, Alt+P pan, Alt+Z zoom, Alt+L length and a key for every tool, plus Alt+Shift for the panels (Alt+Shift+G DICOM tags, Alt+Shift+M Info, Alt+Shift+P Inspect, Alt+Shift+N Annotations, Alt+Shift+H Hang). Alt+/ lists them all. The single-letter keys typed into a report that had the focus; whoever prefers them can turn them back on in the list. Toolbar tooltips name each shortcut.
- 3 viewports side by side, and 3 planes. A 3-viewport layout (Alt+3, or Hang › Layout), and 3 planes for CT and MR (Alt+Shift+3): the exam's own axial, coronal and sagittal series side by side, each on the middle of its series (a plane the exam doesn't have stays empty; MPR is still on its own button).
- Fills in the Annotations list. Each fill made with the Fill tool is listed with its area and mean, to find, rename or delete, and Clear All now clears fills on every viewport with the measurements (one undo brings them back). A result's measurements (an AI's) stay until removed from Results. A fill's numbers can be dragged aside, with any tool, and a line points back to the fill; clicking a fill again makes it one fill with one label, where it used to stack a second label on the first.
- What's downloading, file by file. While the viewer downloads an exam from an archive, the files received are listed under the progress bar: series, image, frames and size.
- Special tools at the end of the toolbar. Spine and Fill moved after the everyday tools and actions; US Extract and Lung show only for an exam with ultrasound.
- View Images on Transfers. A row's right-click menu opens the exam's images straight away, without opening its record.
- Transfers shows when an exam is waiting to complete. A row used to stay blue, "Receiving", from its first image until the completion timeout ran out, long after the last image was in. Now it is blue only while images are arriving (a large image still coming in counts). Once they stop, the row glows a soft yellow and its pill reads Waiting with a live countdown to the Inactivity Timeout (Settings › RadCore › Completion Detection), after which the exam is taken as complete and routed. An image that arrives meanwhile turns the row blue and starts the countdown again. The Dashboard and the transfer bar at the foot of the other pages show the same two states and the same countdown: the Dashboard's recent exams (which used to call a quiet exam "Queued") and a new Waiting line in Live Pipeline, and the bar's Receiving in blue and Waiting in yellow. RadFetch priors still short of their expected images stay Receiving: they wait for the count, not the clock.
- Organs asks before it starts. Outlining a series can take minutes, so the viewer now asks first ("Outline this series?") wherever an outline is started: the toolbar, a viewport's button, the right-click menu and the Organs panel. Outlines already made, or under way, show at once with no question.
- Transfers shows only what you may do. Someone who may look at Transfers but not manage it no longer sees Resend, Resend with Rule, Reprocess OCR and SR, Pin or Delete in a row's right-click menu, nor Forward, Resend, Cancel Forward, Pin and Delete in the exam's details. The server already refused them; now they aren't offered.
- Delete on Transfers' right-click menu. The last item, under a divider, with the same question as Delete in the exam's details. Needs Transfers manage.
- Not for diagnostic use. A small note beside the viewer's series strip says the viewer is for non-diagnostic image review (the same wording as the patient CD viewer).
- The viewer's Summary shows much more. Info in the viewer's top bar now reads the header of the image on screen and follows it as you scroll: patient (with age, sex and issuer of ID), study (date and time, referring physician, institution, how many series and images, study UID), series (number, body part, protocol, series UID), the image (instance, image type, SOP class, acquisition time, size, frames, pixel spacing, slice thickness and location, bits, photometric interpretation, header window, rescale, compression and its UID, SOP Instance UID), acquisition settings (kVp, mAs, CTDIvol, kernel, field strength, TR / TE, sequence, view, contrast), equipment (manufacturer, model, station, serial, software) and what sent it (AE title and software). Rows the header doesn't have are left out; UIDs, the accession and the patient ID copy with a click (the top bar now shows the patient ID too, and its accession and ID copy the same way), and Copy puts the whole summary on the clipboard as text, for a ticket. On a phone it covers the images.
- New tool cursors in the viewer. Each tool's cursor is now an arrow, whose tip is exactly the point the tool acts on, with the tool's own toolbar icon in a small badge beside it, all the same size. They stay sharp on high-resolution screens and show on dark and bright images alike. The Ellipse tool also has its own icon now, so it no longer looks like Circle.
- Picking one day on Query/Retrieve. The study date has a One day | Range switch. One day, the default, is a single date box: pick the day and the search covers that day, with no second date to set. Range shows From and To as before, and a start picked after the end moves the end with it. The choice is remembered.
- Easier to see a series still loading in the viewer. The small spinner at the top of each viewport is replaced by a glowing strip down its right edge that fills as the series loads, from the image shown outward, the way the viewer loads it. On the viewport you're working in, the scroll slider's track shows it instead, with the percentage under it, so you can see how far you can scroll. When the series finishes the strip lights up and fades.
- The image viewer runs on the newest Cornerstone3D (5.17). It opens images stored as JPEG XL, which it couldn't before, and handles touch on tablets and phones better. Everything else draws as before: every compression RadFuzion sees (JPEG, JPEG-LS, JPEG 2000, HTJ2K, RLE), color and cine ultrasound, breast tomosynthesis, MPR and 3D. MPR's crosshairs no longer show up as six measurements in the Annotations list. The patient CD viewer is unchanged.
- Organs has its own panel. The organ outlines' controls and their long list of structures moved out of the Annotations panel into an Organs panel beside the images, like Info and Inspect, opened by Organs in the viewer's top bar (and by outlining a series). Annotations stay readable while the outlines are on.
- Pixel PHI's text scan leaves the server room. A scan uses at most half the server's cores, so routing and the viewer stay quick while it reads an exam.
Fixed
- The transfer bar no longer shows a sent exam as Receiving for a moment. Between a forward ending and the exam list being read again, the bar fell back to what the list had said while the exam was arriving.
- A reformat headed by one image of another plane hangs in its own plane. Some scanners put a single axial image at the front of a coronal or sagittal series. The viewer went by the first image, so 3 planes and the hanging protocol took the series for axial. A series' plane is now the one most of its first five images lie in.
- Cloud retrieves: fewer missed images, and a reason when one is missed. Retrieving an exam from a Google Cloud, Azure DICOM or DICOMweb store could skip images without saying why, then fetch the whole exam again the slow way and lose the count of images to expect. Three causes are addressed. On PostgreSQL with the download concurrency at 100 or more, recording an image asked for a second database connection while still holding one, so every pooled connection ended up waiting and images that waited 15 seconds were skipped (and everything else that needed the database stalled); recording an image now uses one connection at a time, and a retrieve files at most 32 images at once whatever the concurrency setting. A store that answered "too many requests" or "unavailable" lost that image at once; it is now asked twice more, after the wait the store asks for. And every skipped image now says why in the log (the store's HTTP status, or the error), where it used to say nothing. An event-log line that can't be written no longer fails the image it describes.
- Delete removes a cached exam's images too. Deleting an exam (its details' Delete, and now Transfers' right-click menu) removed its record and its files in incoming, forwarded and failed, but left its images in the image cache until the cache evicted them. They are deleted with the rest.
- A damaged image can no longer stop RadFuzion. Every compressed image the server decodes (JPEG, JPEG-LS, JPEG 2000, HTJ2K, RLE) is now decoded in a separate worker process: forwarding uncompressed to a destination (by a rule, or because it doesn't take the compression), the viewer's frames for tomosynthesis and large clips, thumbnails, DICOMweb frames and OCR. A damaged image used to be decoded inside the service, where a codec crash on it would have stopped routing along with everything else; now only that image fails, with the reason, and a fresh worker takes the next one. Images decode exactly as before.
- One image that won't decode no longer holds back the exam. Sending an exam to a destination that doesn't take its compression, one image whose pixel data won't decode stopped the whole batch, so none of the images reached it and the exam kept failing. That image now fails on its own with the reason ("won't decode"), and the rest are delivered; a destination that takes the compression still gets it as stored. The same goes for a workstation retrieving the exam from the image cache: a C-MOVE stopped there too, and a C-GET waited for an answer that never came.
- Full Backup downloads again; the alerts history and the restart check open. These checked a permission that was never set for them, so they refused everyone, admins included, and the backup download then also failed while writing the ZIP. All three work for the roles that should have them (Settings manage; Settings view for the alerts history).
- A retry sends only what's still missing. When an exam reached a destination only in part, each retry sent the whole exam again, so the destination got the images it already had once a try (five times over, for an exam with one image it refuses). A retry now sends each destination only what it lacks, and the images every destination has count as sent, so Retry also sends just the rest.
- No red "Low Fidelity" label in the viewer's 3D view. Turning the MPR 3D view showed Cornerstone's "Transient Low Fidelity" label in red while it drew quick rough frames, which looked like an error. It's gone; whether an image is lossy compressed is still in the viewer's own image text.
- Exams opened from Query / Retrieve hang right at once. While an exam pulled into the viewer cache loaded, each view showed a plain preview picture, and only the active view's went away: a mammogram opened from Query / Retrieve showed its other views unflipped and centered until the pointer passed over each one (opened from Transfers it hung right away). Each view's preview now goes as soon as that view draws, and it's flipped and placed like the image meanwhile.
- Tomo plays every view without the mouse over it. Toolbar Tomo could leave a view on its first slice until the pointer passed over it: only the viewport under the pointer kept its images loading while cine ran. Every playing view now loads and moves, and the image text (Img 12 / 70) keeps up while they play.
- The viewer's 3D cursor does only its own job. Dragging with it also ran the tool picked before it (window/level, a measurement, Fill). Now the other viewports go to the slice through the point, a crosshair marks it on each, and series the point isn't in (a scout, another exam) stay where they are.
- Every series shows its thumbnail. An exam opened from RadFuzion's own storage showed a picture only for the series already in a viewport; the rest showed their modality until opened. Now each one shows from the start.
- Overlay text and lines are white, not purple. Text and lines a modality stores with an image (a dose sheet's text, a scanner's marks) showed purple in the viewer's thumbnails and in exported pages; they're white now, as on the images.
- The viewer's Probe shows the value on every image. Past a series' first image (a CT's slice 2 onwards, every tomosynthesis slice) the Probe drew only a small circle: the Cornerstone3D tool read the wrong image. It now reads the pixel under it, in HU on CT, and its value is listed in Annotations.
- Boxes that take several values let you type them again. Typing a comma, a space or a new line in a box that takes a list was undone the moment it was typed, so a second value couldn't be added: Query/Retrieve's Allowed Calling AE Titles and C-MOVE Destinations ("RADFUZION_812, RADFUZION_DEV"), the DICOM receiver's Also Accept Called AEs and SOP Class UIDs, HTTPS's Additional Certificate DNS Names, RadSR's Never Read These Image Types and workflow filters ("CHEST PA"), RadList's Allowed Calling AE Titles, Modalities and Station Names, RadRead's Excluded Modalities, a Mirth output's channel IDs, and Email notification recipients. Values can be separated by commas or new lines in all of them.
- The viewer's messages show. Messages the viewer gives (a series it can't show, a result that belongs to another series) went unseen; they now show at the top of the images.
- The viewer's pixel value is in HU. The value shown under the pointer on CT (and other rescaled images) had the rescale applied twice, so 160 HU read as -864. Measurements and ROI statistics were right.
- Opening a report kept the viewer as it was. Opening a structured report from the viewer reloaded the images behind it and dropped any segmentation drawn on them.
- X-rays no longer show as negatives in the viewer. Radiographs stored as MONOCHROME1 (common on CR and DX units) opened inverted in the viewer while their thumbnails were right, and Reset did the same. They now show the right way round, including after Reset, and Invert still works.
- The viewer lets go of files it loaded for itself. Files the viewer read on its own (to count a multiframe's frames, or to read a shutter again) stayed in the browser's memory until the page closed, so a long reading session kept growing. It now keeps the newest 32 (512 MB at most).
- Add-ons stop with RadFuzion. If the service stopped unexpectedly, an add-on's work (a text scan, organ outlines, a registration, an install) ran on without it, holding the server's cores and its files. It now stops with the service, on Windows and on Linux.
Security & privacy
- The server's calls need their page's permission. About half of the server's routes didn't check the signed-in user's permission, so anyone signed in who called them directly could, for example, list every exam, download its images, read or delete reports, export or push studies, or read the TLS settings. Each call is now held to the permission of the page it serves (what every signed-in user needs, such as their own account, stays open to them). Opening images in the viewer is for anyone who can open the viewer (Transfers, Query / Retrieve, RadRead, Prelim Reporting or RadShare); the full tag window still needs Tag Editor view. The Dashboard permission shows the Dashboard's recent exams and health.
- The menu shows only the pages a role can open, and a role without the Dashboard starts on its first page instead of an empty Dashboard. A RadRead Tech viewing a report no longer takes its edit lock, which could keep the radiologist out of it.
- The event log and the routing log need their permission. Any signed-in user who called them directly could read them; they now need Troubleshooting view (Settings view reads the event log's HL7 events only, for the HL7 Messages page), like the Troubleshooting page. The event log's CSV export already did. Its live updates, and transfer progress, now reach only those who may see them.
- Search and the page's banners show only what a role can open. The header search lists exams only for Transfers or the Dashboard and log events only for Troubleshooting; the services banners need Troubleshooting or Settings, and the transfer ticker Transfers or the Dashboard. RadHub's list of sites to control shows only to the hub's admins, the only people the hub lets control a site.
- Editing tags in the viewer needs Tag Editor manage. The viewer's tag edits were accepted from any signed-in user who called them directly; they now need the Tag Editor manage permission, like the Tag Editor page. Reading an image's full tag list (the Tags window and Inspect) needs Tag Editor view.
- Add-ons install exactly what was tested. Each add-on carries the list of every package it installs, at an exact version, with each file's SHA-256: an install takes those files or stops. A new release of a library an add-on uses can't change or break it, a changed file is refused, and nothing is built from source on the server.