Curriculum/Pillar 8 · Diagnostic Reasoning in EMG

Pattern-Based Diagnosis

Module 32 of 36·30 min readREAL EMG DATA

Learning objectives

1Operationalize the core electrodiagnostic dichotomies
2Apply localization logic across the peripheral nervous system
3Use real recordings to anchor pattern recognition

The needle examination does not produce a diagnosis; it produces evidence, and the diagnosis is the output of an explicit reasoning engine that the expert runs over that evidence. That engine has three moving parts: a set of dichotomies that convert raw findings into mechanism, a localization logic that uses spared muscles to falsify candidate lesions, and a Bayesian frame that combines the pretest probability set by the clinical question with the likelihood ratios carried by each finding. Pattern-based diagnosis is therefore not pattern matching — it is hypothesis-driven inference, and making its logic explicit is what allows the same waveform to mean different things in different clinical contexts without contradiction.

REAL EMG DATAHealthy, myopathic, and neurogenic recordings side by side

Timebase

ms/div

Healthy control

Crisp tri/biphasic MUAPs around 0.5–1 mV that recruit smoothly as effort rises.

Myopathy

Small, short-duration, often polyphasic units; a dense (early/full) recruitment pattern at low force.

Neuropathy

Giant, long-duration MUAPs (>3 mV) from collateral reinnervation; reduced recruitment with rapid firing.

Feature	Healthy	Myopathic	Neuropathic
MUAP amplitude	Normal (0.5–1 mV)	Low (small units)	High (giant units)
MUAP duration	Normal	Short	Long
Recruitment	Normal	Early / full	Reduced
Firing rate at low force	Low	Low (many units)	High (few units)
Polyphasia	<15%	Increased	Increased

Three real concentric-needle recordings scrolling together: the healthy control (moderate, smooth units), the myopathic muscle (small, short, low-amplitude units with early/full recruitment), and the neurogenic muscle (giant, long-duration reinnervated units with reduced recruitment). Read them as the engine's training set — the contrast between dense-and-small (myopathic) and sparse-and-giant (neurogenic) is the visual anchor for every dichotomy that follows.

Operationalizing the core dichotomies

The engine's first task is to compress a high-dimensional recording into a few binary decisions, each of which carries mechanistic weight. Two dichotomies dominate. The axonal vs demyelinating distinction, drawn from nerve conduction, asks whether the lesion has killed axons (low CMAP/SNAP amplitude, secondary mild slowing, with denervation on needle EMG) or stripped myelin (marked conduction slowing, conduction block, temporal dispersion, with relatively preserved amplitudes distal to the block). The myopathic vs neurogenic distinction, drawn from needle EMG, asks whether the lesion has hollowed motor units from within (small, short, polyphasic units, early/full recruitment) or subtracted them from above (large, long, polyphasic units, reduced recruitment).

These dichotomies are resolved not by single findings but by diagnostic pairings that yoke a discriminating measure to a shared one. The amplitude × recruitment pairing is decisive: tall units with sparse (reduced) recruitment is neurogenic; small units with dense (early/full) recruitment is myopathic. The spontaneous-activity × MUAP-size pairing disambiguates fibrillations — a finding shared by active denervation and irritable myopathy — because fibrillations with large units indicate chronic active neurogenic disease, whereas fibrillations with small units indicate an active (inflammatory, necrotizing, or dystrophic) myopathy. Each pairing forces an ambiguous finding into an unambiguous interpretation by reading it against its discriminating partner.

A side-by-side comparison

Decision variable	Healthy	Myopathic	Neurogenic
MUAP amplitude	Normal (~0.5–1 mV)	Low (small units)	High (giant, >3 mV)
MUAP duration	Normal	Short	Long
Polyphasia	<15%	Increased	Increased
Recruitment	Normal (ratio ~5)	Early / full	Reduced
Firing rate at low force	Low	Low (many units)	High (few units)
Spontaneous activity	None	± fibrillations (irritable)	Fibrillations / PSWs (active)
Discriminating pairing	—	Small units × early recruitment	Large units × reduced recruitment

The table is the engine's lookup, but its rows are not equally weighted. Polyphasia is shared by both disease patterns and never discriminates; the load-bearing variables are amplitude, duration, and recruitment, read in the pairings above. The firing-rate row encodes the most reliable single discriminator at low effort: a few units firing fast betrays neurogenic loss, while many units firing slowly for trivial force betrays myopathic weakness.

Findings falsify hypotheses; they do not merely match patterns

The deep move in expert electrodiagnosis is to treat each muscle as a test of a hypothesis, not as a data point to be catalogued. Before the needle goes in, the examiner holds candidate localizations; each muscle sampled is chosen because its result will falsify at least one of them. A normal muscle is as informative as an abnormal one — often more so — because sparing a muscle that a candidate lesion must involve eliminates that candidate outright. This is hypothesis-driven sampling, and it is why a targeted ten-muscle study can localize more precisely than an undirected twenty-muscle one.

Localization as an explicit falsification process

Once a process is established as neurogenic, localization proceeds by intersecting the territories of the abnormal muscles and, crucially, by using the sparedmuscles to exclude levels. The logic is anatomical and deductive. If a hand muscle is abnormal but the C8–T1 paraspinals are normal, the lesion is unlikely to sit at the root and is pushed distally toward plexus or peripheral nerve, because a root lesion severe enough to denervate the limb muscle should also denervate the paraspinals supplied by the same segment's posterior ramus. Conversely, abnormality confined to muscles of a single peripheral nervedistalto a branch point, with sparing of that nerve's more proximal muscles, localizes to the nerve below the spared branch. The decision is always the same shape: which candidate level is incompatible with this pattern of involvement and sparing?

Worked as an algorithm, localization asks for each candidate lesion: enumerate the muscles it must involve and those it must spare; then test whether the observed map of abnormal-and-normal muscles is consistent. Radiculopathy requires abnormality in a myotomal distribution spanning multiple peripheral nerves plus paraspinal involvement; a plexopathy spans multiple nerves and roots without paraspinal involvement; a mononeuropathy is confined to one nerve's territory distal to the lesion. Each falsification narrows the posterior set until one localization survives.

The Bayesian frame: pretest probability, likelihood ratios, posterior

The reasoning engine is, formally, a Bayesian updater. The pretest probabilityis set by the clinical question before any needle is inserted: a referral for "rule out ALS" in a patient with progressive painless weakness establishes a very different prior over diagnoses than "rule out carpal tunnel" in a typist with nocturnal hand tingling. Each electrodiagnostic finding then acts as a likelihood ratio — fibrillations in three limbs and the tongue carry a large positive likelihood ratio for a diffuse motor-neuron process; a normal study of a clinically weak muscle carries a likelihood ratio that can substantially lower the probability of a neurogenic cause. The posterior is the updated localization and diagnosis, and it is the product of prior and evidence, not the evidence alone.

This frame explains two facts that confound the novice. First, the same finding means different things in different contexts: a single fibrillation potential is weak evidence in a high-prior ALS referral and is more likely artifact or incidental in a low-prior carpal tunnel study. Second, a study must be tailored to its question: the examiner extends the study precisely where the posterior remains uncertain, adding muscles whose results carry the likelihood ratios most capable of separating the surviving hypotheses. Pattern-based diagnosis done well is Bayesian inference made physical — pretest probability from the clinic, likelihood ratios from the needle, posterior localization from their product.

Clinical Pearl

Decide your hypotheses before the first insertion, and let the clinical question set the prior. The examiner who walks in knowing the three localizations worth distinguishing — and which muscle's result would falsify each — runs a shorter, more decisive study than the one who samples broadly and interprets afterward. Sample the spared muscles you expect to be normal as deliberately as the abnormal ones: a well-chosen normal muscle eliminates a candidate lesion and sharpens the posterior more than yet another confirmatory abnormal finding.

Common Pitfall

Do not read findings without their prior. A single fibrillation potential, a borderline-large unit, or mild slowing is a likelihood ratio, not a diagnosis, and its meaning is set by the pretest probability of the clinical question. Treating an isolated finding as decisive regardless of context — calling denervation from one fibrillation in a low-prior study, or dismissing a normal limb muscle as uninformative in a high-prior one — abandons the Bayesian frame that makes electrodiagnosis valid and converts a reasoning engine into a pattern-matcher that confirms whatever it expects.

Key points

Two dichotomies convert findings to mechanism: axonal vs demyelinating (from NCS) and myopathic vs neurogenic (from needle EMG); resolve them with the amplitude × recruitment and spontaneous-activity × MUAP-size pairings.
Polyphasia never discriminates myopathic from neurogenic; amplitude, duration, and recruitment do — and a few units firing fast (neurogenic) vs many firing slowly (myopathic) is the key low-effort discriminator.
Localization is explicit falsification: each muscle tests a hypothesis, and spared muscles exclude levels — paraspinal sparing pushes a lesion distal to the root; confinement to one nerve's distal territory localizes the nerve.
Radiculopathy = myotomal abnormality across multiple nerves WITH paraspinals; plexopathy = multiple nerves/roots WITHOUT paraspinals; mononeuropathy = one nerve distal to the lesion.
Diagnosis is Bayesian: the clinical question sets the pretest probability, each finding is a likelihood ratio, and the posterior localization is their product — so the same finding means different things in different contexts.