BrainGate Point-and-Click Cursor (Kim et al., 2011)
Measured by Kim, Simeral, Hochberg, Donoghue, Friehs & Black · IEEE Trans. Neural Syst. Rehabil. Eng. 19(2) (2011)
Inputs
The measured or assumed values behind the calculations, each with its source.
- A = 278 px
- Mean center-to-target distance across the eight radial targets: 300 px horizontal, 255 px vertical and 278 px diagonal/overall mean.
- W = 48 px
- Target diameter in the radial point-and-click task.
- MT = 7.20 s
- S3 average successful target-acquisition movement time across four sessions.
- success = 97.4 %
- S3 successfully acquired 97.4% of targets across 193 runs; all misses were timeouts, with no false target selections.
- targets = 8
- Eight-target center-out point-and-click task plus a random-target Fitts metric task adapted from HCI pointing-device evaluation.
Strictest ITR
Each scoring method is an upper bound on the channel, so the headline is the strictest (smallest) one for this entry. Use the score selector on the home page to view any single method across entries.
-
Compute the radial-task index of difficulty
ID = log2(A/W + 1) = log2(278/48 + 1) = 2.76 bits
The paper reports the target diameter and center-to-target distances, but not endpoint scatter for an ISO effective-width calculation. This uses nominal width, then discounts for completion failures.
-
Compute successful-run pointing throughput
TP_success = 2.76 bits / 7.20 s = 0.384 bits/s
Movement time is defined only for successful acquisitions in the paper.
-
Discount for incomplete runs
TP_net = 0.384 bits/s x 0.974 = 0.374 bits/s
S3 missed 2.6% of runs by timeout and made no false target selections, so completion rate is applied as a simple net-throughput discount.
What counts as a bit depends on the action space. The number of distinguishable actions and how likely each one is are design choices of the task, not the sensing hardware. The same modality can present a fixed set of targets, a set pruned per step by a grammar or language model, or a continuous control space. Each of these changes how many actions are live and how the probability mass is spread, and therefore the information per selection. Read the action space below before comparing headline numbers across entries.
Action space
What the user can produce at each step, and how those options are distributed.
- Structure
- Continuous control space
- Size
- Continuous
- Prior
- Uniform: all actions assumed equally likely
- Notes
- A 2D neural cursor with a decoded click state, moving from the 2006 proof-of-concept toward a mouse-like computer-control channel. The participant could move the cursor, hold it still and click target regions. This entry uses a Fitts-style calculation from the published radial-task geometry and timing because the task is a continuous point-and-click channel, directly ancestral to later ReFIT cursor and Webgrid-style benchmarks. The Wolpaw-over-8 radial-target figure is kept as supplementary.
Comparability The strictest bound here is Fitts throughput: the index of difficulty, log₂(A/W + 1), per movement. Directly comparable to the other continuous-pointing entries (mouse, trackball, stylus, gaze and the cursor BCIs). Set against the text entries (keyboards, spellers, speech) it crosses methods: both report bits/s, but one measures movement difficulty and the other text information, so compare within the family first.
Other score types
Bounds the atlas keeps out of the default strictest headline: as-reported figures, alternate task conditions, or raw-channel ceilings that shouldn't win the headline by default. Each still carries a score type, so the home-page selector ranks this entry on it when you choose that type. Read its derivation before comparing across entries.
-
Per-selection information with errors
B = log2(8) + 0.974 log2(0.974) + 0.026 log2(0.026 / 7) = 2.75 bits/selection
This is the standard Wolpaw discrete-selection view of the eight cued targets. Because the target is cued and the user drives a continuous cursor to it, this is supplementary rather than the ranking number.
-
Selection rate
60 / 7.20 = 8.33 selections/min
Uses the same S3 four-session average successful-run movement time as the Fitts reference calculation.
-
Information transfer rate
2.75 bits/sel × 8.33 sel/min ÷ 60 s/min = 0.382 bits/s
-
Achieved-bitrate credit per correct acquisition
N = 8 cued targets → log2(N − 1) = log2(7) = 2.81 bits per net-correct selection (field-standard achieved bitrate, e.g. Webgrid; Nuyujukian 2015, which introduced the metric, used log2(N)).
-
Net-correct acquisition rate
incorrect = 0, so net-correct = 0.974 of runs at one per 7.20 s → 0.974 / 7.20 = 0.135 correct/s.
S3 made no false target selections (all misses were timeouts), so there are no false selections to net against. Uses the same S3 four-session successful movement time (7.20 s) and 97.4% completion as the entry's Fitts and Wolpaw calcs. With no false selections, the achieved bitrate credits log2(N − 1) per successful acquisition.
-
Achieved bitrate
2.81 bits × 0.135 correct/s = 0.38 bits/s.
Source
- Authors
- Kim, Simeral, Hochberg, Donoghue, Friehs & Black
- Publication
- IEEE Trans. Neural Syst. Rehabil. Eng. 19(2), 2011
- Reference
- BrainGate publication page