How NOR Discrimination Index Is Calculated

1. Supported workflow with validation-only pose measures

Object-exploration timing is pose-derived (it requires a working pose-estimation backend) and is gated behind an experimental feature flag that is off by default. When it is unavailable -- pose estimation failed, produced no usable evidence, or the feature is disabled -- RodentMetrics reports the discrimination index as unavailable rather than estimating one.

2. What the discrimination index represents

The discrimination index is commonly used in the literature as a measure of an animal's relative exploration preference between two objects -- typically a previously seen ("familiar") object and a newly introduced ("novel") object. RodentMetrics reports it in that same descriptive sense: a number summarizing how exploration time was split between the two objects, computed only from the exploration time it itself confirmed.

It is not a memory test result. RodentMetrics does not interpret the value as evidence of recognition memory, and does not diagnose, score or classify any cognitive state from it.

3. How RodentMetrics identifies object exploration

Object position is defined in the arena editor as one of two shapes -- a rectangle or an ellipse -- placed, sized and rotated per object. Exactly one object must be marked "familiar" and one "novel", and their footprints are validated to stay inside the arena and not overlap.

The animal's position is read from two pose-estimated keypoints, not the tracker's bounding-box center: the nose and the body center. A frame only becomes eligible for a given object when all of the following hold at once:

• both the nose and body-center keypoints are detected with sufficient pose confidence;
• the nose-to-object-boundary distance is within the configured exploration distance (2.0 cm by default, adjustable per experiment);
• the angle between the body-center-to-nose heading and the body-center-to-object-center direction is within 45°, so the animal is oriented toward that object, not merely nearby;
• exactly one of the two objects meets both conditions in that frame -- a frame where both objects are simultaneously eligible is treated as ambiguous and is not attributed to either object.

Meeting this proximity-and-orientation test is only a candidate state, not a confirmed exploration bout. RodentMetrics confirms a bout through one of two evidence paths, each requiring the candidate condition to hold continuously for a short confirmation window (0.125 seconds): an approach path (the nose keeps closing on the object, or touches it) or a local-interaction path (the nose's closest approach reverses or oscillates near the object, or touches it). Simply staying at a stable proximity without any of that motion evidence does not, by itself, confirm a bout.

Once confirmed, a bout tolerates brief orientation or distance fluctuation without ending. It only ends on a confirmed departure -- sustained outward nose movement together with a sustained loss of that local proximity envelope -- or immediately, without any tolerance, on a missing-pose frame, a tracking gap, or the two objects becoming simultaneously eligible (ambiguous). None of these interruption cases are bridged or interpolated; a new bout requires the full candidate-and-confirmation sequence to start again from fresh evidence.

4. Familiar and novel exploration time

Familiar-object time and novel-object time are each the sum of the durations of every confirmed bout attributed to that object. A bout's duration runs from the frame its confirmation window completed to the frame its departure was confirmed (or to the end of the analyzed interval, if the animal is still exploring when the recording ends).

5. The exact RodentMetrics discrimination-index formula

RodentMetrics computes the discrimination index directly from the two confirmed exploration times:

The result is a plain ratio, not a percentage -- it is not multiplied by 100 -- and it is mathematically bounded between −1 and +1. A positive value reflects more novel- than familiar-object exploration time; a negative value reflects more familiar- than novel-object time; zero reflects an even split. These are descriptive facts about the ratio, not a behavioral interpretation.

RodentMetrics also reports a companion novel preference percentage, T_novel ÷ (T_novel + T_familiar) × 100, on the same 0-100% scale researchers may already be used to.

6. Worked example

Take the illustrative, synthetic exploration times below (not data from a real recording):

With 12.0 s of familiar-object exploration and 18.0 s of novel-object exploration (30.0 s total), RodentMetrics would report a discrimination index of 0.200 and a novel preference of 60.0%.

7. Boundary cases and interpretation

A few situations are worth understanding before interpreting a result:

• If only one object ever receives confirmed exploration -- for example 9.0 s of novel-object time and 0 s of familiar-object time -- the discrimination index reaches its mathematical maximum of 1.000 (or minimum of −1.000 the other way), not an error or a special case.
• If neither object receives any confirmed exploration, the discrimination index and novel preference are reported as unavailable, not 0 (Section 5).
• A bout in progress when the analyzed interval ends is closed out at that end time rather than discarded, so exploration right up to the end of the recording is still counted.
• A momentary pose-tracking gap or a frame where both objects are simultaneously eligible ends any exploration in progress immediately and is never bridged; a brand-new candidate-and-confirmation sequence is required afterward, even if the interruption was very short.
• Because Novel Object Recognition's pose-based evidence and detector are experimental, a session can legitimately produce little or no usable evidence -- that is reported as such rather than filled in.

8. What this metric does not establish

The discrimination index summarizes confirmed exploration time already collected. On its own, it does not:

• independently prove that the animal recognized the familiar object as familiar;
• diagnose or measure a cognitive or memory impairment;
• constitute a validated clinical or research outcome measure;
• demonstrate that RodentMetrics' own Novel Object Recognition implementation has been independently validated.

It should be interpreted alongside the underlying exploration times and bout counts, pose-quality information for the session, and the experimental protocol -- not read in isolation as a conclusion about the animal, and not as evidence that the pose-derived detector or measure has been independently validated.

9. What to report

Because the pose-derived exploration detector remains experimental and validation-only, RodentMetrics recommends reporting the underlying parameters alongside the discrimination index itself, so a reader can evaluate the number rather than take it on faith:

• Familiar-object and novel-object exploration time
• The discrimination index, or that it was unavailable due to zero total exploration time
• The novel preference percentage
• Familiar and novel bout counts, and which object was first confirmed
• The exploration distance and orientation-angle settings used
• Pose-quality/coverage information for the session
• That the pose-derived exploration detector and discrimination measure are experimental and validation-only
• The RodentMetrics software version the analysis was run with

Further reading

Antunes M, Biala G. The novel object recognition memory: neurobiology, test procedure, and its modifications. Cogn Process. 2012;13(2):93-110. doi:10.1007/s10339-011-0430-z