eTSA: Endoscopic Transsphenoidal Approaches

• eTSA is a minimally invasive neurosurgical procedure that accesses the sellar region through nasal corridors to resect pituitary adenomas.
• Innovations such as the TACTER robotic system and Cosserat-rod modeling enable precise navigation, achieving sub-anatomical tip accuracy in complex skull base surgeries.
• Live deep learning–augmented image guidance provides real-time, GPS-like feedback to safely navigate critical neurovascular structures during eTSA.

Endoscopic transsphenoidal approaches (eTSA) constitute the prevailing minimally invasive neurosurgical technique for accessing the sellar region at the skull base, including pituitary adenoma resection. The method traverses the nostrils and sphenoidal sinus using endoscopic visualization, demanding precise navigation through confined anatomical corridors adjacent to critical neurovascular structures. Innovations in robotic instrumentation and real-time image guidance have significantly augmented the dexterity, safety, and intraoperative orientation in eTSA, addressing core limitations of rigid tools and static preoperative imaging (Yamamoto et al., 28 Apr 2025, Sarwin et al., 2023).

1. Anatomical Access and Workflow in eTSA

eTSA advances a defined intranasal path to the sella turcica, passing fixed anatomical stations that serve as crucial orientation landmarks. The standard workflow comprises:

• Insertion of the endoscope into the nasal vestibule, navigation along inferior and middle turbinates.
• Mobilization of the nasal septum for enhanced exposure.
• Identification of the sphenoethmoidal recess and sphenoidal ostium.
• Entry into the sphenoidal sinus; removal of septations demarcates the anterior sella.
• Excision of the sellar floor and dural opening.
• Tumor resection within the pituitary fossa, circumspect of the internal carotid artery (ICA), optic nerves, and sellar boundaries.
• Hemostasis and skull base reconstruction (Sarwin et al., 2023).

Key anatomical structures encountered, in order, include septum, turbinates (inferior, middle, superior), choana, sphenoethmoidal ostium, rostrum, sinus septations and floor, sella, planum sphenoidale, cavernous carotid prominences, and the clival recess. Spatial disorientation and rapidly changing intraoperative anatomy (e.g., due to hemorrhage or tumor removal) necessitate continuous re-identification of these stations.

2. Advances in Instrumentation: TACTER Robotic System

Rigid instruments constrain safe navigation around the ICA and cranial nerves. The TACTER system introduces a concentric tendon-actuated continuum robot architecture tailored for eTSA (Yamamoto et al., 28 Apr 2025):

• Architecture: An outer unidirectionally asymmetric notch (UAN) nickel-titanium tube (OD 4 mm) encompasses an independently actuated inner 3D-printed bidirectional segment, both sharing a bending plane but offering 360° spatial access via base rotation.
• Outer robot: Asymmetric notching (radius $r_{01}=1.97$ mm, depth $d=2.93$ mm) enables unidirectional bending through a single $80\,\mu$m tendon.
• Inner robot: Comprised of 25 serial 3 mm OD cylinders housing 0.23 mm nitinol rods (bending spine) and 0.127 mm nitinol tendons for bidirectional curvature; permits translation (up to 30 mm relative to the outer tube) via motorized actuation.
• Follow-the-leader motion: Independent axial translation of the inner tube enables the tip to advance along the preformed curvature of the outer tube, minimizing mucosal trauma and preserving non-linear trajectories critical for circumventing vital anatomy (Yamamoto et al., 28 Apr 2025).

3. Cosserat-Rod Modeling and Control

TACTER’s kinematics and statics are modeled using a two-rod Cosserat framework:

• Kinematics: For each rod $i$, the centerline $P(s)$ and rotation $R_i(s)$ evolve as

$$\dot{P}(s) = R_i(s) v_i(s),\quad \dot{R}_i(s) = R_i(s)[u_i(s)]$$

with $v_i(s)$ (shear/stretch), $u_i(s)$ (curvature/twist).

• Statics: Forces $n_i(s)$ and moments $m_i(s)$ equilibrate as

$$\dot n_i + [u_i] n_i + f_i = 0, \quad \dot m_i + [u_i] m_i + [v_i] n_i + \tau_i = 0$$

with tendon-induced loadings $f_i(s)$ and $\tau_i(s)$ determined by tendon geometry and tension $T_i$.

• Boundary and compatibility conditions: Model solution uses a shooting method to satisfy tip force/moment constraints; the coupled ODEs for $u_1(s)$, $v_1(s)$, $\theta(s)$, $\beta(s)$ yield the tip pose as a function of tendon tensions $(T_1, T_2)$ and inner translation $\Delta z$.
• Input/output: Control inputs are $T_1$, $T_2$, and $\Delta z$; output is the tip 3D position and orientation.

This modeling enables precise prediction of tool tip position with measured mean errors $\leq 5$ mm (2.5% of total length), supporting sub-anatomical accuracy for sensitive skull base work (Yamamoto et al., 28 Apr 2025).

4. Experimental Validation and Clinical Utility

• Benchtop assessment: Across 13 robot configurations and 30 tendon-stroke increments, TACTER’s predicted tip position matched marker-based measurements with $\leq 5$ mm error, worst-case under 12 mm for maximal extension.
• Cadaveric evaluation: Demonstrated navigation from naris to sphenoid sinus with curvature up to $60^\circ$ (outer), $90^\circ$ (inner), accessing the full sellar workspace with minimal mucosal contact and successful delivery of a 635 nm laser fiber.
• Clinical implications: TACTER’s dexterity facilitates trajectories circumventing the ICA, cavernous nerves, and optico-chiasmatic structures while minimizing repetitive shaping and tissue disruption. The 4 mm OD fits existing endonasal corridors and delivers $\geq 2$ DoF curvature at the tip, potentially enhancing gross-total resection rates of tumors with cavernous sinus extension, reducing operative times, and enabling future integration with navigation and sensing modalities (Yamamoto et al., 28 Apr 2025).

5. Intraoperative Navigation: Live Image-Based Guidance

Traditional reliance on preoperative MRI suffers degradation after intraoperative tissue deformation. Sarwin et al. propose deep learning–augmented navigation using live endoscope imagery (Sarwin et al., 2023):

• Real-time object detection: YOLOv7, trained on $3\times10^6$ frames with 19,000 expert-labeled samples, detects 15 anatomical classes plus instrument with mean AP$_{50}$ $= 53.4\%$ (AP$_{50:95} = 26.2\%$), robustly identifying septum (AP$_{50}=78.6$), sella floor (AP$_{50}=70.6$), and sphenoidal sinus (AP$_{50}=74.0$).
• Unsupervised embedding–based roadmap: Sequences of bounding-box detections are encoded via a transformer-based network $E_\theta$ mapping buffered observations to a latent scalar $z_t\in(0,1)$, which monotonically correlates with the anatomical progression along the eTSA corridor.
• Guidance logic: Decoders $D_\text{class}, D_\text{box}$ forecast upcoming or past anatomical landmarks, supporting overlay-based navigation through dynamic environments without metric pre-registration.
• Quantitative correlation: The learned $z_t$ achieves Pearson $r=0.80$ with time index, maintaining strict ordinal alignment with surgical progress.

This real-time, relative-position encoding reflects the endoscopic path as observed, offering “GPS-like” feedback that is independent of shifts in the operative field (Sarwin et al., 2023).

6. Limitations and Future Directions

• TACTER: Integration of real-time closed-loop navigation and electromagnetic or laser-based localization with the tendon-actuated system is proposed for submillimetric targeting. Future work will address greater workspace flexibility and more complex multi-path scenarios (Yamamoto et al., 28 Apr 2025).
• Live roadmap guidance: Presently, the learned coordinate is relative, not an absolute metric. The embedding’s resilience under atypical anatomy or unexpected events (e.g., hemorrhage) requires further validation. Extension to additional neurosurgical approaches and fusion with sparse metric trackers or 3D mapping (e.g., SLAM, depth estimation) is anticipated. Augmentation with other intraoperative modalities, such as fluorescence or Doppler imaging, is proposed. Real-time overlay and alerting for critical structures are feasible at $>$30 Hz throughput with modern workstation hardware (Sarwin et al., 2023).

7. Comparative Summary and Clinical Impact

Advances in eTSA integrate novel continuum robotic platforms and deep-learning–based anatomic guidance. The TACTER device achieves dexterous, minimally traumatic access to the sellar and parasellar regions, validated for both geometric precision and maneuverability. Concurrently, live image-embedded navigation frameworks provide robust intraoperative orientation, overcoming the limitations of MRI-based systems affected by tissue shift and surgical progression. The confluence of these technological trajectories promises enhanced resection rates, reduced morbidity, and an expanded scope for endoscopic skull base surgery (Yamamoto et al., 28 Apr 2025, Sarwin et al., 2023).