Offline Personalization

Summary:

TIP V5.0 introduces an optional offline personalization workflow for users with data privacy requirements. A personalizer pipeline, available as part of Sim4Life, is run locally to generate an anonymized, segmented head model from a T1 MRI scan. The tool deforms facial features beyond recognition while fully preserving the internal tissue structures relevant for TI simulation planning. Only the anonymized, segmented output is uploaded to the cloud for electromagnetic simulation.

---

Motivation

In the standard TIP workflow, magnetic resonance (MR) images are uploaded directly to the cloud for processing, including automatic segmentation from MR images and personalization of the head model. Some users and institutions have concerns about uploading identifiable medical images to remote servers. To address this, TIP V5.0 provides a local anonymization workflow that ensures no personally identifiable anatomical features leave the user's local environment.

How It Works

The anonymization workflow consists of two stages:

1. Local Preprocessing (on your machine)

The local personalizer pipeline (run_personalizer.bat) generates an anonymized, personalized head model with electrode placement from a T1 MRI scan using Sim4Life.

Prerequisites

• Sim4Life version 9.4 or later, installed on Windows.
• A T1 MRI scan in NIfTI format (.nii.gz). The filename must contain t1.
• The personalizer Python package is automatically installed into Sim4Life's bundled Python environment on first run.
• Atlas data (~37 MB) is downloaded automatically to %USERPROFILE%\.personalizer on first run.

Input files

Place your input files in the project's input directory:

project_root\
└── inputs\
    ├── input_1\
    │   └── subject_t1.nii.gz       (required — T1 MRI scan)
    ├── input_2\
    │   └── fiducials.sab            (optional — electrode positions)
    └── input_3\
        └── sim_settings.json        (optional — simulation settings)

Running the tool

Open a Command Prompt and run the batch file. Paths containing spaces must be enclosed in double quotes:

run_personalizer.bat --sim4life "C:\Program Files\Sim4Life_9.4" --project-root C:\data\subj01 --electrode-radius 7.0

Key arguments:

Argument	Description
--project-root PATH	Project root directory (contains inputs/ and outputs/)
--electrode-radius MM	Electrode radius in mm (e.g. 7.0)
--electrode-type TYPE	Preset type: ROUND_1_6CM2 or ROUND_3CM2
--sim4life PATH	Sim4Life installation directory (if non-default)
--atlas NAME	Brain atlas to register (e.g. ICBM_152)
--dti	Process DTI/DWI data for anisotropic conductivity
--register-dti	Register DTI to MRI T1 before conductivity extraction (safe only for minor translations with no rotation — if rotation is involved, register the T1 to the DTI yourself)
--denoise-dti	Denoise DTI data before conductivity extraction
--max-bval INT	Exclude DTI shells above this b-value threshold (e.g. 1000)
--dti-model MODEL	Diffusion model for conductivity extraction: auto, dti, or dki

You must provide either --electrode-radius or --electrode-type.

What it does

• The tool deforms facial features (face, ears, etc.) beyond recognition, removing identifying characteristics from the image.
• Crucially, internal tissue structures relevant for TI planning are not degraded by this process.
• It automatically segments the different tissues in the image and generates a personalized head model with electrode placement.

Output

The pipeline produces the following files in the project's outputs/ directory:

Folder	Contents
outputs/output_1/	SMASH file (.smash) — head model
outputs/output_2/	SAB file (.sab) — head model
outputs/output_3/	T1 image (.nii.gz) — processed T1
outputs/output_4/	Diffusion Tensor (.nii.gz) — DTI in S4L convention
outputs/output_5/	JSON config — target coordinates
outputs/output_6/	SAT file (.sat) — head model
outputs/output_7/	T1 image resampled (.nii.gz) — processed T1
outputs/output_8/	Label field (.nii.gz) — tissue labels
outputs/output_9/	Tissue list (.txt) — tissue names

All output files are bundled into a zip archive (results.zip) along with a personalizer_args.json file that records the exact parameters used.

Tip: To inspect the anonymized head model locally, open Sim4Life, go to File > Open..., and select the .smash file from the outputs.

2. Cloud Simulation (AWS)

• Upload the output zip directly to the TIP platform.
• The platform runs 66 ohmic quasi-static electromagnetic simulations on the AWS cloud using the anonymized personalized head model.
• These 66 basis simulations can be recombined at the postprocessing stage to reproduce the TI fields for any electrode montage — there is no need to rerun simulations when exploring different configurations.

Cost

Running the cloud simulations for a personalized head model costs approximately $60 per model. This cost is at the charge of the user and is managed through the Billing Center.

Key Benefits

• Privacy: Identifiable facial features never leave your local machine. Only the anonymized, segmented head model is uploaded.
• Accuracy: Internal tissue geometry is fully preserved — simulation quality is not compromised.
• Flexibility: The 66 precomputed basis simulations allow postprocessing exploration of any electrode montage without additional simulation cost.
• Integration: The personalizer pipeline is bundled with Sim4Life (version 9.4+) and installs its dependencies automatically on first run.

Requirements

• A local installation of Sim4Life version 9.4 on Windows.
• The offline personalizer batch file: Download.
• A T1-weighted MRI scan in .nii.gz format meeting the data quality requirements.
• A TIP account with sufficient credits for cloud simulation (see Billing Center).