Connectomics & Tractography for Target Selection

Evidence status. Clinical utility is target- and workflow-specific. Tractography is a model derived from diffusion MRI; normative connectomes and patient-specific data have complementary biases. Prospective evidence is scarce and narrow: blinded prospective data support tractography-guided contact selection as non-inferior to clinical testing, without establishing general superiority of connectivity-based surgical targeting.

1.Structural and functional connectivity

Diffusion MRI estimates local fiber orientation. A tracking algorithm joins those estimates into streamlines; a streamline is not an axon and streamline count is not axon count. Deterministic tracking follows a dominant direction and is reproducible but can stop at crossings. Probabilistic tracking samples uncertainty and may recover more plausible branches while also producing more false-positive paths.

The scale is worth internalizing. When many independent groups reconstructed the same simulated dataset with known ground truth, most recovered the majority of the real bundles but also produced more anatomically invalid bundles than valid ones. Sensitivity was high; specificity was the problem, because the false-positive bundles looked just as plausible as the true ones.

Functional connectivity measures correlated signal over time, commonly with resting-state fMRI. It is sensitive to preprocessing, motion, medication, disease state, and parcellation. It does not establish monosynaptic connection or causal direction.

2.Patient-specific versus normative connectomes

Patient-specific imaging can reflect individual anatomy and distortion but often has limited signal-to-noise, especially near implants, lesions, and deep crossings. Normative datasets offer high-quality acquisitions and stable group estimates, but may not represent a patient's age, disease, surgery, or reorganization.

3.Tremor and the cerebellothalamic system

Tractography can model the dentato-rubro-thalamic tract or related cerebellothalamic pathways when planning DBS or focused ultrasound. It may help explain a therapeutic corridor spanning ventrolateral thalamus and posterior subthalamic area. Results depend strongly on seed, waypoint, exclusion masks, tracking direction, b-value, number of directions, and whether decussating fibers are modeled.

Use it beside established landmarks and clinical testing. A tract that reaches cortex through anatomically impossible territory is a pipeline warning, not a novel discovery.

4.DBS sweet spots and symptom networks

Lead localization, volume-of-tissue-activated models, and outcomes pooled across cohorts can identify probabilistic sweet spots or connectivity fingerprints in Parkinson disease, dystonia, OCD, depression, epilepsy, and pain. What that yields depends almost entirely on how much data stands behind it, and these remain population estimates whose performance can shrink when exported to a new center. One multicenter study analyzed 534 electrodes overall, including 394 electrodes in discovery cohorts of 197 bilaterally implanted patients with four disorders, plus additional validation datasets. It identified an orderly distribution of outcome-associated frontal circuits. The strength of validation varied by disorder and modeling approach, and prospective application involved only three patients. These are promising network hypotheses rather than a universal targeting mandate.

Retrospective sweet spots also fail in a specific and instructive way. A response-associated OCD bundle connecting dorsal anterior cingulate and ventrolateral prefrontal cortex with the anteromedial subthalamic nucleus was derived largely from normative connectome data. When Widge and colleagues asked a related question with patient-specific diffusion imaging in eight patients with ventral capsule/ventral striatum DBS, connectomic models correlated with outcome but did not predict it above chance. A small patient-specific cohort that fails to reproduce a finding may reflect cohort, acquisition, or model differences; it limits transportability without proving that either connectome is ground truth.

A useful map specifies the cohort, clinical scale, imaging normalization, stimulation model, validation method, and whether the finding was discovered and tested in independent data.

5.Lesions and disconnection

Connectomic lesion mapping can relate outcomes or adverse effects to networks reached by a lesion. In capsulotomy, cingulotomy, thalamotomy, pallidotomy, and subthalamotomy, this can explain why lesion location alone is incomplete. It cannot retrospectively recover tissue physiology or eliminate confounding by indication and center technique.

6.Validate before interpretation

7.Known failure modes

Crossing, kissing, fanning, and sharply turning fibers create false negatives or false continuations. Partial volume, edema, hemorrhage, atrophy, and postoperative susceptibility can distort orientation. Gyral bias systematically over-represents terminations at gyral crowns and under-represents the sulcal banks and fundi, so a cortical connectivity profile is partly a map of folding geometry. Threshold tuning after seeing the desired tract introduces confirmation bias.

These are not theoretical margins. In a retrospective study of 40 patients (22 subthalamic implants for Parkinson disease and 18 ventral intermediate implants for essential tremor), Deuter and colleagues held the acquisition and the patient constant and varied only the workflow. Changing the binarization threshold produced nonlinear, unpredictable variation in the measured distance from the active contact to the dentato-rubro-thalamic tract of up to 1.72 ± 1.49 mm; manual measurement differed from automated measurement by 0.91 ± 1.36 mm on average and by 14.9 mm in the worst case; and measurement after normalization to MNI space differed from native space by 0.82 ± 0.50 mm (maximum 2.34 mm). The tract did not move. The pipeline did.

8.Use connectivity to change a decision

Before ordering advanced mapping, state the decision it could alter: target, entry, trajectory, contact selection, lesion location, or programming. If no result would change the decision, the map is decorative. If it will change the decision, define anatomic vetoes and verification criteria in advance.

Hold the technique to the standard it asks of everything else. The cited evidence does not establish general superiority of tractography-defined or connectome-defined surgical targeting over conventional targeting across indications. Distinguish surgical lead placement from later contact selection, and explain the indication-specific evidence in consent. The prospective evidence that does exist is narrower than the enthusiasm around it: in a prospective, double-blinded single-center trial of 40 patients and 57 hemispheres, published in Brain Stimulation in 2026, contacts chosen by maximum overlap of the modeled stimulation volume with the patient's own dentato-rubro-thalamic tract were non-inferior to contacts chosen by standardized monopolar review, with point estimates favoring clinical testing in both the tremor and the Parkinson arms. The primary result concerns acute total tremor control within a prespecified non-inferiority margin. It does not establish equivalence for every symptom, durable benefit, or superior surgical targeting; clinical assessment remains necessary.

The long-term standard should be prospective: preserve raw data and transforms, localize the delivered stimulation or lesion, collect blinded outcomes, and test whether connectivity added predictive value beyond routine anatomy and clinical variables.