rsna-anonymizer 19.0.0.dev2__tar.gz → 19.0.0.dev4__tar.gz

{rsna_anonymizer-19.0.0.dev2 → rsna_anonymizer-19.0.0.dev4}/.vscode/settings.json

@@ -11,7 +11,7 @@
     "editor.defaultFormatter": "charliermarsh.ruff", 
     "editor.formatOnSave": true,
   },
-  "python.languageServer": "Default",
+  "python.languageServer": "None",
   "python.analysis.typeCheckingMode": "basic",
   "python-envs.defaultEnvManager": "ms-python.python:poetry",
   "python-envs.defaultPackageManager": "ms-python.python:poetry",

{rsna_anonymizer-19.0.0.dev2 → rsna_anonymizer-19.0.0.dev4}/CHANGELOG.md

@@ -4,6 +4,17 @@ All notable changes to this project will be documented in this file
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [19.0.0.dev4]
+- Fix welcome screen clipped on macOS Retina (Tk 9): size main window after welcome layout
+- Help menu: **AI Features** opens setup (not only the Welcome screen button)
+- Readme: macOS install with `python-tk@3.12`, uv managed Python 3.12 with Tcl/Tk 9, in-venv Tk verification
+
+## [19.0.0.dev3]
+- English AI Features help: overview after Overview, linked tool pages (not in Help menu), in-app `help:` navigation, wider layout
+- AI Features setup opened from Welcome screen only; help and UI messages updated accordingly
+- Remove Burnt-in Annotation help: click green rectangles after Detect Text to whitelist text
+- Readmes document uv-based install and upgrade for faster V19 setup
+
 ## [19.0.0.dev2]
 - AI batch processing (harmonize, face blur, remove pixel PHI) with progress UI and memory guards
 - Batch OCR and MONOCHROME1 blackout aligned with Series View pixel pipeline

{rsna_anonymizer-19.0.0.dev2 → rsna_anonymizer-19.0.0.dev4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: rsna-anonymizer
-Version: 19.0.0.dev2
+Version: 19.0.0.dev4
 Summary: RSNA DICOM Anonymizer
 Project-URL: Homepage, https://github.com/RSNA/anonymizer
 Author-email: Algorhythm Software <michael@dx.life>
@@ -37,7 +37,7 @@ Requires-Dist: totalsegmentator>=2.3.0
 Requires-Dist: xgboost>=2.0.0
 Description-Content-Type: text/markdown
 
-# RSNA DICOM Anonymizer (V18 stable / V19 dev)
+# RSNA DICOM Anonymizer (V19)
 [![de](https://img.shields.io/badge/lang-de-blue.svg)](readme.de.md)
 [![es](https://img.shields.io/badge/lang-es-blue.svg)](readme.es.md)
 [![fr](https://img.shields.io/badge/lang-fr-blue.svg)](readme.fr.md)
@@ -50,49 +50,80 @@ Description-Content-Type: text/markdown
    - Select "Add python.exe to PATH"
    - Enable "tcl/tk and IDLE"
 ### macOS
-1. Install Homebrew if not present: `/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)'
-2. Install Python 3.12 with Tcl/Tk:
-```
-brew install python@3.12
-brew install tcl-tk
+1. Install Homebrew if not present: `/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"`
+2. Install Python **3.12** (not Homebrew's default `python3`, which may be 3.14), tkinter, and OpenMP (required for Harmonize contrast analysis on macOS):
+
+```bash
+brew install python@3.12 python-tk@3.12 libomp
 ```
+
+Homebrew's `python@3.12` does not include tkinter by itself — use **`python-tk@3.12`**.
+
 ### Linux (Ubuntu/Debian)
 1. Install the required packages:
 ```
 sudo apt update
 sudo apt install software-properties-common
 sudo add-apt-repository ppa:deadsnakes/ppa
-sudo apt install python3.12 python3.12-tk
+sudo apt install python3.12 python3.12-tk python3.12-venv
 ```
 ## Verify Installation
+Use **Python 3.12** explicitly (`python3.12`, not `python3`):
+
+```bash
+python3.12 --version
+python3.12 -m tkinter
+python3.12 -c "import tkinter as tk; print('Tk', tk.TkVersion)"
 ```
-python --version
-python -m tkinter
+
+If python + tkinter has been installed successfully a small GUI window should open. On macOS, **Tk should be 9.0** for correct UI rendering with CustomTkinter 6.
+
+## Install rsna-anonymizer from PyPI
+
+V19 is on PyPI as a pre-release (`--pre`). Use [uv](https://docs.astral.sh/uv/) — much faster than plain `pip` for large ML dependencies (PyTorch, TotalSegmentator, etc.).
+
+### One-time: install uv
+
+```bash
+curl -LsSf https://astral.sh/uv/install.sh | sh
 ```
-If python + tkinter has been installed successfully a small GUI window should open
-## Install rsna-anonymizer package from PyPI
 
-Stable (V18):
+Windows (PowerShell): `powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"`
+
+Restart the terminal if `uv` is not found.
+
+On macOS, update uv and install its managed Python 3.12 with Tcl/Tk 9 (recommended when Homebrew Python is broken or unavailable):
 
 ```bash
-pip install rsna-anonymizer
+uv self update
+uv python install --reinstall 3.12
 ```
 
-V19 development pre-release (requires `--pre`; does not replace the stable install above):
+### Create a virtual environment
 
 ```bash
-pip install --pre rsna-anonymizer
+uv venv rsna-anonymizer --python 3.12
+source rsna-anonymizer/bin/activate    # Windows: rsna-anonymizer\Scripts\activate
 ```
 
-Pin a specific dev build:
+Verify **inside the activated venv** (must be Python 3.12.x and Tk 9.0 on macOS):
 
 ```bash
-pip install --pre rsna-anonymizer==19.0.0.dev2
+python --version
+python -c "import tkinter as tk; print('Tk', tk.TkVersion)"
 ```
 
-Verify: `pip show rsna-anonymizer` or `rsna-anonymizer --version`. See [CHANGELOG](CHANGELOG.md#1900dev2) for dev release notes.
+If `uv venv --python 3.12` gives Tk 8.6, run `uv python install --reinstall 3.12` and recreate the venv. If Homebrew Python works, you can use `python3.12 -m venv rsna-anonymizer` and `python -m pip install --pre rsna-anonymizer` instead of `uv venv`.
 
-TotalSegmentator, XGBoost, and related dependencies are included. Enable Harmonize, Face Blur, and Remove Pixel PHI per project in **Settings → Project** (or when creating a new project). Download models and apply the face license from the AI Features panel.
+### Install the package
+
+```bash
+uv pip install --pre rsna-anonymizer
+```
+
+Verify: `uv pip show rsna-anonymizer` or `rsna-anonymizer --version`. See [CHANGELOG](CHANGELOG.md) for release notes.
+
+TotalSegmentator, XGBoost, and related dependencies are included. Download models and accept the face license from **Help → AI Features** or the **AI Features** button on the Welcome screen.
 ## Execution
 `rsna-anonymizer`
 ### Headless Mode
@@ -100,16 +131,10 @@ You need to provide a path to a project configuration to run in headless mode
 `rsna-anonymizer -c path/to/ProjectModel.json`
 ## Upgrading
 
-Stable:
-
-```bash
-pip install --upgrade rsna-anonymizer
-```
-
-V19 dev pre-release:
+Activate the virtual environment first (`source rsna-anonymizer/bin/activate`).
 
 ```bash
-pip install --upgrade --pre rsna-anonymizer
+uv pip install --upgrade --pre rsna-anonymizer
 ```
 ## Documentation
 [Help files](https://rsna.github.io/anonymizer)

rsna_anonymizer-19.0.0.dev4/docs/2_ai features.html

@@ -0,0 +1,69 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>RSNA DICOM Anonymizer AI Features</title>
+    <link rel="stylesheet" href="styles.css">
+</head>
+<body>
+	<h1>AI Features</h1>
+	<p>
+	The Anonymizer can use optional <b>AI Features</b> to improve de-identification beyond removing patient names and IDs from DICOM tags.
+	These tools work on the <b>image pixels</b> or on <b>series descriptions</b> and are designed for research datasets where burned-in text, recognizable faces, or inconsistent series names could still identify a patient.
+	</p>
+	<p>
+	All AI Features are <b>optional</b>. You choose which tools to turn on for your project. None of them run until you enable them and complete any required setup (such as downloading helper files).
+	</p>
+
+	<h2>Opening AI Features Setup</h2>
+	<p>
+	From the <b>Welcome</b> screen, click the <b>AI Features</b> button to open setup.
+	</p>
+	<p>
+	The setup window lists each available tool with a short description, an on/off checkbox, and a status line (for example “Ready” or “Models not downloaded”).
+	</p>
+
+	<h2>First-time setup (all tools)</h2>
+	<ol>
+		<li>On the <b>Welcome</b> screen, click <b>AI Features</b>.</li>
+		<li>Turn on the tools you plan to use by checking the box next to each name.</li>
+		<li>If a <b>Download</b> button appears, click it and wait until the status shows the tool is ready. Downloads are stored on your computer for future sessions.</li>
+		<li>For <b>Face De-identify</b> only: you must also accept the academic license when prompted (see <a href="ai_features/face de-identify.html">Face De-identify</a>).</li>
+		<li>Click <b>Save</b> or close the dialog to keep your choices.</li>
+	</ol>
+	<p>
+	An internet connection is needed only while downloading helper files. After that, the tools run locally on your machine.
+	</p>
+
+	<h2>Available tools</h2>
+	<p>Each tool has its own help page with step-by-step configuration and usage:</p>
+	<ul>
+		<li><a href="ai_features/remove burnt-in annotation.html">Remove Burnt-in Annotation</a> — finds and removes text burned into the image (patient labels on X-rays, ultrasound overlays, etc.).</li>
+		<li><a href="ai_features/harmonize.html">Harmonize</a> — suggests a standard CT series description from anatomy and contrast (RadLex playbook style).</li>
+		<li><a href="ai_features/face de-identify.html">Face De-identify</a> — blurs the face region on head CT scans to reduce the chance of recognition.</li>
+		<li><a href="ai_features/ai batch process.html">AI Batch Process</a> — runs one or more of the tools above on many studies at once from the PHI Index.</li>
+	</ul>
+
+	<h2>Where you can run AI Features</h2>
+	<ul>
+		<li><b>Series View</b> — work on a single series: detect/remove text, harmonize description, or review face blur (see Help/Operation for Series View).</li>
+		<li><b>AI Batch Process</b> — on the PHI Index, select studies and run the enabled tools in bulk (see <a href="ai_features/ai batch process.html">AI Batch Process</a>).</li>
+	</ul>
+
+	<h2>Turning a tool off</h2>
+	<ul>
+		<li>Uncheck the tool in the <b>AI Features</b> setup window (Welcome screen).</li>
+		<li>If you disable a tool whose files were already downloaded, the Anonymizer may ask whether to remove those files from disk to save space.</li>
+		<li>Disabling a tool does not undo changes already applied to exported images.</li>
+	</ul>
+
+	<h2>If something is not ready</h2>
+	<ul>
+		<li>On the <b>Welcome</b> screen, click <b>AI Features</b> and read the status line under each tool.</li>
+		<li>Complete any pending download or license step.</li>
+		<li>On macOS, Harmonize may require an extra system library (<i>libomp</i>) for contrast analysis; install via Homebrew if the status message mentions OpenMP or XGBoost.</li>
+	</ul>
+
+</body>
+</html>

rsna_anonymizer-19.0.0.dev4/docs/ai_features/ai batch process.html

@@ -0,0 +1,79 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>RSNA DICOM Anonymizer AI Batch Process</title>
+    <link rel="stylesheet" href="../styles.css">
+</head>
+<body>
+	<h1>AI Batch Process</h1>
+	<p>
+	<b>AI Batch Process</b> runs one or more enabled AI Features on <b>many studies at once</b> from the PHI Index, instead of opening each series manually in Series View.
+	It is useful after import when you want to harmonize descriptions, remove burnt-in text, or blur faces across an entire cohort.
+	</p>
+
+	<h2>Before you start</h2>
+	<ol>
+		<li>Open a project and import the studies you want to process.</li>
+		<li>On the <b>Welcome</b> screen, click <b>AI Features</b> and enable the tools you need (see <a href="../2_ai features.html">AI Features</a>).</li>
+		<li>Complete downloads and the face license if using Face De-identify.</li>
+		<li>Confirm each enabled tool shows a ready status.</li>
+	</ol>
+
+	<h2>Starting a batch run</h2>
+	<ol>
+		<li>On the dashboard, open the <b>PHI Index</b> (patient/study list).</li>
+		<li>Select one or more studies (or patients) using the same selection methods as for export.</li>
+		<li>Click <b>AI Batch Process</b>.</li>
+		<li>In the options dialog, check each algorithm you want to run:</li>
+		<ul>
+			<li><a href="remove burnt-in annotation.html">Remove Burnt-in Annotation</a> — choose black out or blend mode.</li>
+			<li><a href="harmonize.html">Harmonize</a> — CT series descriptions only.</li>
+			<li><a href="face de-identify.html">Face De-identify</a> — head CT face blur.</li>
+		</ul>
+		<li>Click <b>Start</b>. A progress dialog shows the current study, series, and phase.</li>
+	</ol>
+
+	<h2>Order of operations</h2>
+	<p>When you select multiple tools, the batch runner processes them in this order for each series:</p>
+	<ol>
+		<li>Remove Burnt-in Annotation (if selected)</li>
+		<li>Harmonize (if selected)</li>
+		<li>Face De-identify (if selected)</li>
+	</ol>
+	<p>This order avoids re-reading pixels after they have been modified and keeps memory use predictable on large jobs.</p>
+
+	<h2>What happens during processing</h2>
+	<ul>
+		<li>Each series is evaluated. If a tool does not apply (wrong modality, already done, not ready), it is skipped and noted in the log.</li>
+		<li>You can cancel between series; the current series may finish its active step before stopping.</li>
+		<li>A summary at the end lists completed, skipped, and failed items.</li>
+		<li>Detailed lines are written to the project log file for troubleshooting.</li>
+	</ul>
+
+	<h2>After batch completes</h2>
+	<ul>
+		<li>Spot-check representative series in Series View or View Index projections.</li>
+		<li>Export anonymized data when satisfied (see Help/Operation).</li>
+		<li>Re-running batch on the same studies usually skips work already recorded in project metadata (for example pixel-PHI already scanned), unless you clear caches or force reprocessing.</li>
+	</ul>
+
+	<h2>Common messages</h2>
+	<ul>
+		<li><b>No AI features are enabled</b> — on the Welcome screen, click <b>AI Features</b> and turn on at least one tool.</li>
+		<li><b>AI features are not ready</b> — open <b>AI Features</b> from the Welcome screen and finish model downloads or license steps.</li>
+		<li><b>Skipped (not CT)</b> — Harmonize or Face De-identify was selected but the series is another modality.</li>
+		<li><b>Skipped (already harmonized / already scanned)</b> — previous batch or manual run recorded completion; safe to ignore unless you intentionally want to redo work.</li>
+	</ul>
+
+	<h2>Related help pages</h2>
+	<ul>
+		<li><a href="../2_ai features.html">AI Features</a> — enable tools and downloads</li>
+		<li><a href="remove burnt-in annotation.html">Remove Burnt-in Annotation</a></li>
+		<li><a href="harmonize.html">Harmonize</a></li>
+		<li><a href="face de-identify.html">Face De-identify</a></li>
+	</ul>
+
+</body>
+</html>

rsna_anonymizer-19.0.0.dev4/docs/ai_features/face de-identify.html

@@ -0,0 +1,74 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>RSNA DICOM Anonymizer Face De-identify</title>
+    <link rel="stylesheet" href="../styles.css">
+</head>
+<body>
+	<h1>Face De-identify</h1>
+	<p>
+	On some <b>head CT</b> scans, facial features in the reconstructed images could allow someone to recognize a patient even after DICOM tags are removed.
+	<b>Face De-identify</b> finds the face region in the scan, blurs those voxels, and saves updated images suitable for research sharing.
+	</p>
+
+	<h2>What you need before using it</h2>
+	<ol>
+		<li>On the <b>Welcome</b> screen, click <b>AI Features</b>.</li>
+		<li>Check <b>Face De-identify</b>.</li>
+		<li>Click <b>Download</b> for the face segmentation model if prompted.</li>
+		<li>Accept the <b>academic license</b> when prompted (required by the segmentation software vendor). This is a one-time acknowledgement on your machine.</li>
+		<li>Save your settings.</li>
+	</ol>
+
+	<h2>Which studies qualify</h2>
+	<ul>
+		<li><b>Modality:</b> CT only.</li>
+		<li><b>Anatomy:</b> Head or head-and-neck coverage where a face segmentation is feasible.</li>
+		<li>Very small fields of view, corrupted stacks, or non-diagnostic scouts may be skipped with a message explaining why.</li>
+	</ul>
+
+	<h2>How it works (in plain terms)</h2>
+	<ol>
+		<li>The tool converts the CT series into a 3D volume.</li>
+		<li>It runs a face segmentation model to locate voxels belonging to the face.</li>
+		<li>It blurs those voxels using the blur mode you selected (see below).</li>
+		<li>Blurred slices are written as a new DICOM series (or update path used by your workflow) while preserving geometry and unrelated anatomy.</li>
+		<li>Segmentation masks are cached so re-running on the same series is faster.</li>
+	</ol>
+
+	<h2>Blur modes</h2>
+	<ul>
+		<li><b>Gaussian blur</b> — smooth blur inside the face mask (common default).</li>
+		<li><b>Mean / median blur</b> — alternative smoothing styles; choose in Series View or batch options if available.</li>
+	</ul>
+	<p>After batch or automatic blur, open the <b>Face blur review</b> dialog when offered to confirm coverage on sample slices before export.</p>
+
+	<h2>Using Series View (one series)</h2>
+	<ol>
+		<li>Open <b>Series View</b> for a qualifying head CT series.</li>
+		<li>Run face blur from the series actions (when enabled and ready).</li>
+		<li>Review axial slices in the review dialog: check that the face is covered and that critical anatomy outside the mask is unchanged.</li>
+		<li>Accept or re-run with <b>force</b> if you changed inputs and need a fresh segmentation.</li>
+	</ol>
+
+	<h2>Batch processing many studies</h2>
+	<p>See <a href="ai batch process.html">AI Batch Process</a>. Select <b>Face De-identify</b> in the batch dialog. Ineligible series are skipped and counted in the batch summary.</p>
+
+	<h2>Limitations and tips</h2>
+	<ul>
+		<li>Designed for CT head data — not for plain radiographs or MRI facial sequences.</li>
+		<li>Always review a subset of blurred series; segmentation can miss unusual positioning or extreme anatomy.</li>
+		<li>Blurring reduces recognizability but is not a guarantee against all re-identification risk.</li>
+		<li>Face model download and license acceptance require internet once; processing itself is local.</li>
+	</ul>
+
+	<h2>Related help pages</h2>
+	<ul>
+		<li><a href="../2_ai features.html">AI Features</a></li>
+		<li><a href="ai batch process.html">AI Batch Process</a></li>
+	</ul>
+
+</body>
+</html>

rsna_anonymizer-19.0.0.dev4/docs/ai_features/harmonize.html

@@ -0,0 +1,77 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>RSNA DICOM Anonymizer Harmonize</title>
+    <link rel="stylesheet" href="../styles.css">
+</head>
+<body>
+	<h1>Harmonize</h1>
+	<p>
+	Research datasets often use inconsistent <b>Series Description</b> names for similar exams (“CHEST WO”, “CT thorax”, etc.).
+	<b>Harmonize</b> analyzes a <b>CT series</b> and suggests a standardized description based on which body regions are visible and whether IV contrast appears to be present.
+	Descriptions follow the RSNA Radiology Playbook (RadLex) style, for example combining regions like <i>Chest+Abdomen</i> and contrast such as <i>IV Contrast</i>.
+	</p>
+	<p>
+	<b>Important:</b> Harmonize only updates the <b>SeriesDescription</b> DICOM tag. It does not change the pixel data.
+	</p>
+
+	<h2>What you need before using it</h2>
+	<ol>
+		<li>On the <b>Welcome</b> screen, click <b>AI Features</b>.</li>
+		<li>Check <b>Harmonize</b>.</li>
+		<li>Click <b>Download</b> for anatomy segmentation models if prompted (this may take several minutes and requires internet once).</li>
+		<li>On macOS, install <i>libomp</i> via Homebrew if the status mentions XGBoost or OpenMP.</li>
+		<li>Save your settings.</li>
+	</ol>
+
+	<h2>Which studies qualify</h2>
+	<ul>
+		<li><b>Modality:</b> CT only.</li>
+		<li><b>Geometry:</b> Standard axial (or suitable) multi-slice volumes work best. Scout images, MIP series, or very short stacks may be skipped or produce limited results.</li>
+		<li>The series must already be imported into your project.</li>
+	</ul>
+
+	<h2>How it works (in plain terms)</h2>
+	<ol>
+		<li>The tool builds a 3D view of the CT series.</li>
+		<li>It runs anatomy segmentation to see which structures (lungs, liver, brain, etc.) are present.</li>
+		<li>It estimates whether the exam used IV contrast (contrast phase analysis).</li>
+		<li>It combines region and contrast information into a suggested <b>SeriesDescription</b> string.</li>
+		<li>Results are cached on disk so repeating Harmonize on the same series is faster unless you force a refresh.</li>
+	</ol>
+
+	<h2>Using Series View (one series)</h2>
+	<ol>
+		<li>Open <b>Series View</b> for a CT series.</li>
+		<li>Click <b>Harmonize Description</b>.</li>
+		<li>A dialog shows the suggested description, body regions found, and contrast information.</li>
+		<li>Choose <b>Apply</b> to write the new description to the anonymized series metadata, or cancel to leave tags unchanged.</li>
+	</ol>
+
+	<h2>Batch processing many studies</h2>
+	<p>See <a href="ai batch process.html">AI Batch Process</a>. Select <b>Harmonize</b> in the batch dialog. For each eligible CT series, the tool computes or reuses a harmonized description and applies it automatically.</p>
+
+	<h2>Clearing cached results</h2>
+	<ul>
+		<li>If you re-import slices or change a series, cached harmonize data may be stale.</li>
+		<li>Series View and batch processing can force re-analysis when needed; cache folders live under each series directory in storage.</li>
+	</ul>
+
+	<h2>Limitations and tips</h2>
+	<ul>
+		<li>Harmonize is a suggestion tool — review descriptions before publishing a dataset.</li>
+		<li>Non-CT modalities (MR, US, XR) are not supported.</li>
+		<li>First run on a large study takes longer; later runs use cached segmentation where possible.</li>
+		<li>Harmonize does not replace clinical reporting or protocol naming from the original site.</li>
+	</ul>
+
+	<h2>Related help pages</h2>
+	<ul>
+		<li><a href="../2_ai features.html">AI Features</a></li>
+		<li><a href="ai batch process.html">AI Batch Process</a></li>
+	</ul>
+
+</body>
+</html>

rsna_anonymizer-19.0.0.dev4/docs/ai_features/remove burnt-in annotation.html

@@ -0,0 +1,70 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>RSNA DICOM Anonymizer Remove Burnt-in Annotation</title>
+    <link rel="stylesheet" href="../styles.css">
+</head>
+<body>
+	<h1>Remove Burnt-in Annotation</h1>
+	<p>
+	Some images contain <b>burnt-in text</b> — patient names, dates, hospital names, or measurements drawn directly onto the pixels (common on ultrasound, CR/DR, and some CT/MR screenshots).
+	Standard DICOM de-identification removes tags in the file header but leaves this visible text in the picture.
+	</p>
+	<p>
+	<b>Remove Burnt-in Annotation</b> uses optical character recognition (OCR) to find text in the image and replace it with a solid black region or a blended patch, depending on your chosen mode.
+	</p>
+
+	<h2>What you need before using it</h2>
+	<ol>
+		<li>On the <b>Welcome</b> screen, click <b>AI Features</b>.</li>
+		<li>Check <b>Remove Burnt-in Annotation</b>.</li>
+		<li>Click <b>Download</b> if shown, and wait until the status is ready.</li>
+		<li>Save your settings.</li>
+	</ol>
+
+	<h2>How it works (in plain terms)</h2>
+	<ul>
+		<li>The tool reads each image frame the same way Series View displays it (including window/level for grayscale images).</li>
+		<li>It searches for text-like regions and compares them against rules and optional whitelists (allowed words such as “L” or “R” for left/right markers).</li>
+		<li>Matched text is covered using either <b>black out</b> (solid rectangle) or <b>blend into background</b> (softer fill).</li>
+		<li>Changes are written to new DICOM files in your anonymized storage; originals in the PHI area are not modified by batch runs until you export the updated series.</li>
+	</ul>
+
+	<h2>Using Series View (one series at a time)</h2>
+	<ol>
+		<li>Open a project and go to the PHI Index or View Index, then open <b>Series View</b> for a series.</li>
+		<li>Click <b>Detect Text</b> to highlight regions the tool thinks are text. Detected regions appear as <b>green rectangles</b> on the image.</li>
+		<li>To <b>keep</b> text inside a green rectangle (for example “L”, “R”, or a slice number you want to preserve), <b>click the rectangle</b>. That text is added to the whitelist on the left and will not be removed when you run <b>Remove Text</b>.</li>
+		<li>Use <b>Remove Text</b> to apply removal on the current frame or series according to the active mode.</li>
+		<li>You can draw <b>blackout rectangles</b> manually for areas OCR missed (toolbar in Series View).</li>
+		<li>Modality-specific <b>whitelists</b> (under project locale assets) define text that should be kept, such as orientation markers.</li>
+	</ol>
+
+	<h2>Removal modes</h2>
+	<ul>
+		<li><b>Black out text</b> — fills detected text with black (or the appropriate background for inverted grayscale). Best when you want obvious, uniform removal.</li>
+		<li><b>Blend into background</b> — tries to match surrounding pixels. Useful when a hard black box would look unnatural.</li>
+	</ul>
+	<p>When running <b>AI Batch Process</b>, you choose the mode in the batch options dialog before processing starts.</p>
+
+	<h2>Batch processing many studies</h2>
+	<p>See <a href="ai batch process.html">AI Batch Process</a>. Select <b>Remove Burnt-in Annotation</b> in the batch dialog. The tool runs on each slice of each selected series where OCR finds text.</p>
+
+	<h2>Limitations and tips</h2>
+	<ul>
+		<li>Very small or low-contrast text may be missed; use manual blackout rectangles if needed.</li>
+		<li>Short numeric strings (slice numbers, etc.) may be ignored by design to avoid false positives.</li>
+		<li>Always spot-check a sample of images before releasing a dataset.</li>
+		<li>OCR models require a one-time download; they do not send your images to the internet during processing.</li>
+	</ul>
+
+	<h2>Related help pages</h2>
+	<ul>
+		<li><a href="../2_ai features.html">AI Features</a></li>
+		<li><a href="ai batch process.html">AI Batch Process</a></li>
+	</ul>
+
+</body>
+</html>

{rsna_anonymizer-19.0.0.dev2 → rsna_anonymizer-19.0.0.dev4}/docs/index.html

@@ -13,11 +13,12 @@
         <ol>
             <h1>RSNA DICOM Anonymizer</h1>
             <li><a href="1_overview.html">Overview</a></li>
-            <li><a href="2_deidentification protocol.html">De-identification Protocol</a></li>
-            <li><a href="3_managing anonymization projects.html">Managing Anonymization Projects</a></li>
-            <li><a href="4_project settings.html">Project Settings</a></li>
-            <li><a href="5_operation.html">Operation</a></li>
-            <li><a href="6_license.html">License</a></li>
+            <li><a href="2_ai features.html">AI Features</a></li>
+            <li><a href="3_deidentification protocol.html">De-identification Protocol</a></li>
+            <li><a href="4_managing anonymization projects.html">Managing Anonymization Projects</a></li>
+            <li><a href="5_project settings.html">Project Settings</a></li>
+            <li><a href="6_operation.html">Operation</a></li>
+            <li><a href="7_license.html">License</a></li>
         </ol>
     </div>
     

{rsna_anonymizer-19.0.0.dev2 → rsna_anonymizer-19.0.0.dev4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "rsna-anonymizer"
-version = "19.0.0.dev2"
+version = "19.0.0.dev4"
 description = "RSNA DICOM Anonymizer"
 readme = "readme.md"
 license = { text = "RSNA-MIRC Public License" }