Distinguishing Neurite Continuity Through Image and Morphology Analysis

Good to see you again. In the previous lesson, you learned the mechanics of a controlled local trace: establish an anchor in the 2D EM imagery, move one section at a time, keep local landmarks in view, and use the 3D mesh only for orientation. That procedure answers where did the profile go?

This lesson addresses the harder interpretive question: did it continue as the same neurite, or did I accidentally begin following a neighboring neurite? This distinction is at the heart of reliable proofreading. A segmentation color, a smooth-looking 3D mesh, or a plausible-looking single image is not enough. You will build a conclusion from continuity across sections, membrane boundaries, local geometry, and the morphology of the larger branch.

For now, remain in no-edit inspection mode. You are learning to make a defensible observation before later lessons introduce edits.

---

The central idea: continuity is a hypothesis tested over depth

In a single EM section, two nearby neurites can be nearly indistinguishable. Both may be narrow, both may contain similar internal texture, and both may run in roughly the same direction. A profile that looks like a perfect continuation in one slice can turn out, a few slices later, to belong to a different process.

So do not ask, “Which nearby profile looks most similar?” Ask:

Which candidate preserves the most consistent physical path through the consecutive sections?

A likely continuation should be supported by several kinds of evidence at once:

Evidence	What you look for	Why it matters
Membrane continuity	A membrane-bounded region remains traceable from slice to slice	This is your primary evidence
Local position	The profile shifts by a plausible amount and direction	Abrupt, unexplained jumps are suspicious
Shape evolution	Width, orientation, and outline change gradually across slices	Cross-sections naturally change as a process bends
Neighborhood continuity	The target retains a coherent relation to nearby structures	Context helps resolve similar-looking candidates
Branch geometry	The candidate fits the incoming branch’s trajectory	A useful, but secondary, anatomical constraint
3D morphology	The selected path forms a plausible local reconstruction	Good for orientation; insufficient by itself

Think of this as a debugging process. The segmentation gives one proposed object identity; the EM stack is the evidence you use to test that proposal. A single visual clue is like a passing unit test: reassuring, but not enough to validate the whole system.

Strong versus weak evidence

Not every clue deserves equal weight.

Strong evidence is visible over multiple consecutive sections:

• a continuous membrane boundary;
• a profile that can be tracked without unexplained spatial jumps;
• a branch that maintains a coherent physical route through the volume;
• two processes that are visibly separate despite being close together.

Weak evidence may guide your attention, but cannot establish identity alone:

• the same segmentation color;
• similar darkness or texture in a single slice;
• apparent proximity in a zoomed-out 3D mesh;
• an assumption that the straightest route must be correct;
• a biological expectation about what the cell “ought” to do.

The last point matters. Morphology is valuable, but it is a constraint, not a substitute for image evidence. Neurites bend, branch, narrow, travel alongside one another, and can temporarily look counterintuitive in a 2D section.

---

A practical decision protocol: follow, compare, verify

When a neurite approaches a confusing region, use the following protocol rather than making an immediate visual guess.

1. Trace confidently up to the ambiguous point

Start several slices before the uncertainty, where the target is unmistakable. Identify:

• the incoming neurite’s center and membrane boundary;
• its approximate direction of travel;
• its width or shape over several sections;
• one or two neighboring landmarks.

This creates a baseline. If you begin at the ambiguous junction itself, you have no reliable history of what the branch was doing before it reached that point.

2. Name the candidate continuations

On the next slice, there may be one apparent successor or several. Mentally label them:

• Candidate A: the most spatially direct continuation;
• Candidate B: a nearby profile that could be a different neurite;
• Candidate C: another plausible profile, if present.

Naming candidates prevents a common failure mode: once your eye chooses one profile, you unconsciously stop evaluating alternatives.

3. Trace each plausible candidate for several sections

Do not decide at the first difficult slice. Follow Candidate A forward for a short run, then return to the last certain slice and follow Candidate B through the same range.

For each candidate, check whether the evidence accumulates or collapses:

• Does a membrane-bounded process remain visible?
• Does the candidate preserve a plausible local displacement?
• Does it remain in the same neighborhood relative to distinctive structures?
• Does it develop into a branch path that makes geometric sense?
• Does the apparent connection vanish once you move two or three slices away?

A neighboring neurite often looks plausible at exactly one crossing or contact point, but fails when traced beyond it. The true continuation may initially look less obvious yet becomes consistently supported across the stack.

4. Inspect the 3D view as a consistency check

Once the 2D evidence favors one route, inspect the corresponding 3D reconstruction at a reasonably close zoom.

Ask:

• Does the selected branch make an abrupt kink, detour, or implausible crossing?
• Does it appear to jump from one nearby process to another?
• Is there a thin unexplained protrusion or disconnected-looking region?
• Does the 3D view support the 2D conclusion, or merely look smooth because the mesh hides the local ambiguity?

Return to 2D if the mesh raises doubt. The 3D mesh is rendered from the segmentation being evaluated, so it can display a segmentation error smoothly.

---

What normal continuity looks like

A true continuation does not have to look identical from slice to slice. A neurite is a three-dimensional tube-like structure, while each EM image is a two-dimensional section through it. Its profile can change substantially as the neurite changes angle relative to the slicing plane.

A likely continuation often has these characteristics:

1. A bounded local displacement. The profile moves through the image in a coherent way rather than disappearing and reappearing far away without contextual explanation.

2. Gradual morphological evolution. A small round profile may become larger and elongated, then narrow again. A branch may appear to rotate. These changes are normal if they unfold continuously over adjacent sections.

3. Persistent neighborhood. Nearby structures may shift too, but their overall arrangement gives a local frame of reference. For example, the target may remain beside the same larger process for several sections.

4. A compatible trajectory. If the incoming neurite trends diagonally through several slices, a candidate continuing in a broadly compatible direction is more plausible than one that requires a sharp change. Treat this as supporting evidence, not proof.

5. Consistency in both directions. Trace a few sections forward from the ambiguous point, then return and trace backward. The same physical interpretation should remain plausible in both directions.

A true branch versus a mistaken switch

A legitimate neuronal branch creates one incoming process and two outgoing continuations. The key question is whether both outgoing paths visibly arise from the same membrane-enclosed structure over successive sections.

At a genuine branch point, you should be able to find a small region in depth where:

• the parent process broadens or changes shape;
• the two daughter paths become distinguishable;
• the membrane geometry supports a physical bifurcation.

If one supposed daughter instead belongs to a separate membrane-bounded neurite that merely passes nearby, it is not a branch. It is a neighboring process.

---

The main traps: crossings, parallel neighbors, and path swaps

The most misleading cases occur when neurites are densely packed, cross in projection, or run alongside each other.

Perpendicular crossings: apparent intersection is not connection

In 2D, a neurite running left-to-right can overlap a second neurite running roughly up-and-down. At the crossing slice, the image may resemble an X, a T, or an ambiguous thickened region. The segmentation may even assign both paths to one object.

To distinguish a continuation from a crossing neighbor:

• Trace the incoming neurite several slices before the encounter.
• At the encounter, examine whether its boundary remains separate from the crossing process.
• Move through several slices beyond the contact.
• Compare the two possible paths: the incoming branch’s original direction versus the crossing neurite’s direction.
• Look for a plane difference: one process may pass above or below the other, becoming visible as you step through depth.

The correct continuation frequently maintains the incoming process’s general trajectory. But direct membrane evidence across several sections remains more important than geometric neatness.

Parallel neighbors: the most dangerous visual look-alike

Two neurites can run close together in the same direction for a considerable distance. In this situation, either one can look like the other’s continuation. They may touch briefly, making the segmentation join them, or a small image artifact may cause the segmentation to jump from one neurite to the other.

The danger is a path swap: the reconstructed path begins on one physical neurite and then continues along its neighbor.

Warning signs of a possible path swap include:

• the path abruptly changes width, texture, or local direction;
• one neurite seems to end while a very nearby parallel process suddenly becomes selected;
• the expected route continues visibly as a separate, unselected process;
• the 3D reconstruction has a suspicious kink, spur, discontinuity, or crossing;
• the current segmentation follows an apparently reasonable path but fails direct membrane tracing.

The FlyWire proofreading guide calls attention to path swaps and emphasizes both 3D visual breaks and the directionality of a continuation.

Internal landmarks: useful, but contextual

Within a neurite, dark oval mitochondria can be distinctive landmarks. If you see a mitochondrion inside a membrane-bounded branch before an ambiguity, then find a plausible continuation whose membrane appears to surround the corresponding structure after it, that supports continuity.

However:

• a mitochondrion alone is not the neurite;
• nearby processes may contain similar organelles;
• organelle visibility changes between sections;
• segmentation may handle the mitochondrion separately from the surrounding neurite.

Use internal landmarks together with the enclosing membrane and the branch’s path. The reliable question is not “Did I find a dark oval?” but “Can I trace the same membrane-enclosed neighborhood through this region?”

---

When image quality disrupts continuity

Sometimes the problem is neither morphology nor segmentation. One image section may be shifted relative to neighboring sections, damaged, poorly imaged, folded, or obscured. In such a case, every profile in the field can appear to jump.

A global shift has a useful signature: many nearby structures jump together in approximately the same direction. A single-neurite mismatch is more likely to be a local tracing or segmentation issue.

If you suspect an image misalignment:

1. Stop trying to infer identity from the target alone.
2. Zoom out enough to inspect multiple neighboring structures.
3. Find a distinctive landmark visible on both sides of the disrupted section.
4. Compare the target’s position relative to that landmark before and after the jump.
5. Record the site as affected by an image artifact if the evidence remains weak.

The useful distinction is:

Observation	More likely interpretation
Only the selected path changes abruptly; neighbors remain coherent	Potential false continuation or path swap
Many structures shift together between two slices	Image misalignment or acquisition artifact
A clear membrane separates two close profiles over several slices	Neighboring neurites, not one continuation
One membrane-bounded process visibly forks into two paths	Likely true branch
The target cannot be followed across a damaged region	Uncertain case; do not force a connection

At this stage, uncertain is a valid and often high-quality result. The next lesson will formalize how to classify reconstruction problems. Your present responsibility is narrower: state whether the visual evidence supports one continuation over another.

---

A 15-minute inspection drill

Use any safe exploratory site containing several nearby processes. Do not add, remove, or otherwise edit segments.

Pass 1: establish the incoming path

1. Choose a neurite that is unambiguous for at least three adjacent sections.
2. At the last clear slice before a crowded region, record an anchor coordinate.
3. Note its approximate direction, width, and one nearby landmark.
4. Keep the segmentation overlay translucent enough that membrane boundaries remain visible.

Pass 2: compare alternatives

1. Move one section at a time into the difficult region.
2. Identify every plausible successor, not just the one carrying the selected color.
3. Trace the first candidate forward for three to five sections.
4. Return to the anchor and trace the second candidate for the same range.
5. Compare which candidate preserves continuity of membrane, position, context, and geometry.

Pass 3: check morphology and document the result

Rotate the 3D mesh at close range, then return to your 2D anchor. Record one of these evidence-based outcomes:

site: [x, y, z]
incoming profile: [brief shape and landmark description]
candidates considered: A [description], B [description]

likely continuation: A / B / unresolved
2D evidence: [membrane, displacement, landmarks across N sections]
3D consistency: [supports / raises concern / uninformative]
confidence: high / moderate / low
edits: none

A concise note such as “A remains membrane-continuous for six sections; B is a parallel neighbor separated by a visible boundary in three sections” is much stronger than “A looked right.”

---

Takeaways and next step

You can now distinguish a likely continuation from a nearby neurite by treating the problem as evidence accumulation across the EM stack:

• Start before the ambiguity and trace from a confident anchor.
• Compare all plausible candidates rather than committing to the first visually convenient one.
• Prioritize membrane continuity and multi-slice tracking.
• Use position, local landmarks, width, and branch direction as supporting evidence.
• Treat 3D morphology as a consistency check, never as proof that overrides 2D EM.
• Be especially cautious at perpendicular crossings, parallel bundles, and apparent path swaps.
• Recognize image-wide jumps as possible alignment artifacts.
• Leave a case unresolved when the evidence does not support a confident identification.

Next, you will turn these observations into an explicit classification: whether a suspicious case is a false split, false merge, missing branch, or genuinely uncertain.