Getting started

Requirements

• macOS 26.5 or later
• A folder containing a WFDB record (.hea + .dat) you want to view — PhysioNet’s MIT-BIH Arrhythmia Database is the canonical test set.

Install

Murmur Studio is distributed through the Mac App Store. Install it via the App Store listing and launch it from Applications or Spotlight.

Loading a record

On launch the window shows the full bedside shell in its idle state — every surface present, waiting for a record. Two ways in, both in the File menu:

• File ▸ Open Record… (⌘O) opens the system file picker. The same action is inline in the empty window (“Open Record Folder…”) and in the toolbar’s ⋯ menu.
• File ▸ Open Recent — folders you’ve opened before reopen with one click. Sandbox-safe: each entry stores a security-scoped bookmark, not a raw path.

Once a folder is open:

1. The left sidebar lists every record found — record name, signal count, sample rate, and duration.
2. Click a record to import. The first import on a record runs the WFDB decoder + min/max pyramid + manifest writer. Subsequent visits load instantly from the cache.

The app is sandboxed, so the file picker deliberately asks for a folder rather than a single file — that way the security scope covers the .hea, its sibling .dat, and any annotation files. WFDB multi-frequency records (per-signal .dat files with format[xspf] suffixes) feed straight in; no separate ingest step.

Corpora with a RECORDS index

Large PhysioNet corpora are not one flat folder of .hea files. WFDB’s convention is a RECORDS index at the root listing record names or subdirectories, each of those carrying its own RECORDS — PhysioNet’s ECG Arrhythmia Database is 45,152 records across 452 leaf folders that way.

Pick the corpus root. Murmur reads the index and lists the whole corpus, with each row’s identity its path relative to the folder you picked. A folder with no RECORDS is scanned exactly as before, so a MIT-BIH directory behaves identically.

Two things worth knowing:

• The index is the corpus author’s own statement of what the corpus contains, so it is read rather than second-guessed — a .hea sitting in the tree that RECORDS does not list was excluded on purpose, and Murmur honours that.
• A scan of that size takes seconds and reports its progress in the window while it runs. If the index names a record whose .hea is missing, or a path that leaves the folder you granted, the count is reported when the corpus opens — the list is never quietly short.

Records import lazily, one click at a time, so opening a 45,000-record corpus costs one scan, not 45,000 imports.

Reading the bedside view

The stage reads top to bottom in the order you navigate: whole record → hour → window.

• Variability metrics strip (Murmur Pro) — time- and frequency-domain HRV and QT variability for the whole record, the current window, or the hour, with a scope picker. Every number states its provenance.
• Overview and hour bands — the full recording’s envelope with the viewport marked, and a 60-minute band beneath it. Click either to jump; drag to scrub.
• The paper — a Metal-rendered ECG grid with the trace and its annotation layers. Standard View pins the clinical calibration (25 mm/s · 10 mm/mV); gain and speed presets (5/10/20 mm/mV, 25/50 mm/s) are one click away, and the seam beneath them always discloses the actual on-screen scale. A lock holds calibration across zoom gestures so paging a long record can’t drift the paper.
• The docked inspector — the calibration readout and the beat card. Hover the trace to focus a beat; click to pin it. With Murmur Pro the card reads per-beat PR, QRS, QT, and QTc (selectable correction formula) against the patient’s own normal template, with calibrated uncertainty stated on every QT.
• Lead chips — one chip per lead above the stage. Click a chip to add or remove that lead from the overlay; ⌥-click to show one lead alone. The first lead in selection order is the primary: marks, the docked inspector, and the calibration readout all follow it, and overlaid reads carry a “Measured on …” attribution so a number is never ambiguous about its source lead.

• Lead presets — the chip bar’s presets menu applies a named set of leads in one click: Limb, Precordial, Bipolar limb, All leads, plus anything you save. Presets are stored by lead name, not by channel, so a set you build on one record applies to every record that carries those leads.

  Each row says how much of itself the open record can satisfy — Precordial — 3 of 6 — before you click. A built-in set that none of this record’s leads can satisfy is omitted from the menu; a preset you saved is always listed, disabled rather than silently doing nothing (#351). Matching ignores case, but nothing else: Murmur does not map one lead spelling to another, because WFDB sources do not agree on a closed set of lead names and an alias table would be an equivalence the record never stated.

If the record carries low-rate trend channels — anything below 5 Hz: HR, SpO₂, etCO₂, tidal volume, state probabilities, alarm flags — they render in their own strips below the ECG canvas, time-locked to the same viewport: vitals sparklines, one lane per boolean alarm channel, a ventilation-state backdrop, and a signal-quality heat band.

With Murmur Pro, the trend stack below the stage adds computed whole-record lanes — beat-derived heart rate, RMSSD, LF/HF, and the QTc interval trend — on one shared time axis. Click a lane to move the trace to that moment. Lanes toggle from View ▸ Trend Lanes.

The context drawer under the stage carries the record’s documents: comments from the .hea header, your anchored notes, and (with Murmur Pro) the morphology section — whole-record beat clustering with representative shapes, where the algorithm clusters and the analyst classifies.

Navigation

Gesture	Action
Drag chart left/right	Pan all channels in time-lock
Pinch / ⌘-scroll	Zoom around the center
Zoom presets (24 h … 2 s)	Set the window width directly
Click overview or hour band	Jump to that point
J / K	Next / previous finding
← →	Pan one window
Click a finding or queue row	Center the viewport on it

Findings and annotation layers

Findings reach the canvas three ways:

• The record’s own annotator layer (e.g. a WFDB .atr file) renders automatically — beat badges on the trace, categories in the review queue.
• The in-app arrhythmia scan (Murmur Pro) proposes ranked VT/VF candidate episodes and rhythm-event candidates (AF, pauses), each with its score, span, and model provenance.
• External producers can drop a JSON file named <recordName>.annotations.json next to the record’s .hea — see the annotation schema for the wire format. Range findings render as translucent fills, point findings as thin rules at the sample index.

Reviewing and triaging

The review queue (toolbar toggle, or View ▸ Show Review Queue) lists every finding and candidate, grouped and filterable by category and confidence. Unlock editing from the toolbar lock icon, then each row exposes the disposition controls:

• Confirm (✓) — record the finding as VT, VF, or “confirmed (unsure).”
• Dismiss (✗) — mark it a false positive.
• Reset (↶) — clear the disposition back to unreviewed.

Tally chips track confirmed / dismissed / to-review counts. Dispositions persist inside the imported bundle, so re-running an upstream producer never overwrites your review. Notes, endorsements, and authored findings (Murmur Pro) save into a portable session document (⌘S), and confirmed findings can be exported as WFDB annotations or a citable report from the Export menu.

Murmur Pro

One in-app purchase (Settings ▸ Purchases) unlocks every instrument above marked (Murmur Pro), current and future: beat calipers, variability metrics, computed trend lanes, the arrhythmia scan, morphology clustering, and annotation authoring. The viewer — import, calibrated paper, annotation layers, review dispositions, exports — stays free.