Robotic Needle Insertion for CT-guided Percutaneous Biopsy of Thoracoabdominal Lesions: A Prospective Multicenter Randomized Trial

Abstract

Purpose

To compare safety and feasibility between a novel CT-guided robotic system and the conventional freehand technique for puncture biopsy of thoracoabdominal lesions.

Materials and Methods

In this prospective multicenter randomized trial, individuals with suspected lesions were enrolled between July 2023 and April 2024 across three university teaching hospitals and randomized to the robot-assisted group (n = 82) or the freehand group (n = 83). Procedure outcomes included the technical success rate, targeting error, number of CT scans and needle adjustments, puncture time, and complications. Descriptive and inferential statistics were calculated.

Results

A total of 165 participants (mean age, 60 years ± 10 [SD]; 83 male) were included. Compared with the freehand group, the robot-assisted group demonstrated a higher technical success rate (97.56% [80 of 82] vs 62.65% [52 of 83], P < .001), lower targeting error (mean Euclidean deviation: 1.7 mm ± 1.1 vs 4.5 mm ± 3.9, P < .001), and fewer CT scans (mean, 4.3 ± 1.9 vs 5.2 ± 2.3; P = .002) and needle adjustments (mean, 0.7 ± 0.7 vs 1.6 ± 1.6; P = .003). Despite differences in geometric precision, both groups achieved 100% (82 of 82 and 83 of 83) diagnostic yield. The median puncture time was comparable between groups (5.5 minutes ± 4.3 vs 4.8 minutes ± 7.0, P = .50). During lung biopsies, the robot-assisted approach yielded fewer complications compared with the freehand approach (4.88% [four of 82] vs 16.87% [14 of 83], P = .014).

Conclusion

Compared with the freehand approach, robot-assisted biopsy yielded greater precision and reduced adjustments and complications while demonstrating noninferior diagnostic efficacy and comparable duration.

Keywords: Robotic Needle Insertion, Biopsy, Thoracoabdominal Lesions, Robot-assisted Biopsy, CT-guided Intervention, Percutaneous Needle Biopsy, Randomized Controlled Trial, Algorithm Development, CT, Clinical Testing, Interventional-Body, Biopsy/Needle Aspiration, Percutaneous, Thorax, Abdomen/GI, Liver, Lung, Kidney

©RSNA, 2026

Summary

In a multicenter randomized trial, a novel robotic system for CT-guided percutaneous thoracoabdominal biopsy improved needle placement accuracy and technical success while reducing complications, compared with the conventional freehand approach.

Key Points

• ■ In this prospective multicenter randomized trial, CT-guided robot-assisted biopsy was shown to be feasible and safe while improving the technical success rate for thoracoabdominal lesions, compared with the freehand technique (97.56% [80 of 82] vs 62.65% [52 of 83], P < .001).

• ■ Compared with freehand biopsy, robotic assistance resulted in lower targeting error (mean, 1.7 mm ± 1.1 [SD] vs 4.5 mm ± 3.9; P < .001) and fewer CT scans (mean, 4.3 ± 1.9 vs 5.2 ± 2.3; P = .002) and needle adjustments (mean, 0.7 ± 0.7 vs 1.6 ± 1.6; P = .003), but a comparable mean puncture time (5.5 minutes ± 4.3 vs 4.8 minutes ± 7.0, P = .50).

• ■ Compared with the freehand group, the robot-assisted group had fewer complications from lung biopsy (4.88% [four of 82] vs 16.87% [14 of 83], P = .014).

Introduction

Image-guided percutaneous diagnostic and therapeutic procedures, which include biopsies and ablation, are widely adopted for minimally invasive interventions, as evidenced by the growing market and technological advancements (1,2). During these procedures, the proceduralist generally places the needle manually via guidance from imaging modalities such as US, CT, or MRI (3–5). CT is a versatile modality for image-guided procedures because of its fast volume acquisition, high resolution, and widespread availability. Traditionally, however, these procedures are performed freehand and thus are dependent on the experience of the operator. During particularly challenging procedures, the operator needs to adjust the direction of the needle with a substantial frequency; each time, CT scans are required to assess the needle tip position, resulting in long procedure times, additional radiation exposure, potential complications, or nondiagnostic biopsy samples.

Although numerous commercialized navigation systems, including Maxio (Perfint Healthcare) (6), Robio Ex (Perfint Healthcare) (7), InnoMotion (Innomedic) (8), iSYS (iSYS Medizintechnik) (9), RC 120 (Shanghai Simple Touch Technology) (10), ANT-C (NDR Medical Technology) (11), Epione (Quantum Surgical) (12), and the Imactis CT electromagnetic navigation system (GE HealthCare) (13), have been designed to assist in percutaneous interventions with procedure trajectory planning, each presents inherent limitations. CT fluoroscopy provides real-time anatomic feedback but exposes both the operators and the patients to substantial doses of radiation (14). Cone-beam CT integrates intraoperative three-dimensional imaging into the interventional suite; however, compared with conventional CT, it provides inferior soft tissue contrast, and it is highly susceptible to metal and respiratory motion artifacts (15). Electromagnetic navigation does not involve intraoperative radiation, but it is susceptible to electromagnetic interference from metallic equipment, registration errors caused by dynamic respiratory motion, and the inability to accurately monitor needle deflection in real time (16). Consequently, there remains a critical need for a robotic assistance system that combines the high spatial resolution of conventional CT with rigid mechanical guidance—ensuring precise needle placement—and the primary robotic benefit of minimal radiation (17).

Advanced robotic systems for percutaneous puncturing, such as the light puncture robot (18), Zerobot (Medicalnet Okayama) (19), AcuBot (Johns Hopkins) (20), and ACE Robotic System (XACT Robotics) (21), have been developed to not only position and orient a needle guide or holder on the basis of preprocedural CT images but also insert the needle automatically.

However, studies that have provided evidence on the effectiveness of such systems are characterized by certain limitations, including single-center study designs, limited sample sizes, and evaluations conducted exclusively on phantom models. We hypothesized that, compared with the conventional freehand technique, robot-assisted needle insertion would achieve noninferior diagnostic efficacy while providing fewer needle placement deviations, fewer procedural adjustments, and a lower complication rate for CT-guided percutaneous biopsy of thoracoabdominal lesions. Therefore, this study introduces a table-mounted system that is specifically engineered to plan a precise puncture trajectory and achieve intraoperative navigation and automatic step-by-step needle insertion; subsequently, the physician can manually biopsy the tissue with remarkable accuracy. The purpose of this work is to further explore the safety and feasibility of the robotic system and to compare it with the conventional freehand method for thoracic and abdominal percutaneous puncture under CT guidance through a prospective multicenter randomized trial.

Materials and Methods

Study Design and Participants

This prospective randomized trial was conducted with the approval of the institutional review boards of the three participating university teaching hospitals (center 1: Fudan University Shanghai Cancer Center, approval no. 2305275-3-2406B; center 2: Zhejiang Cancer Hospital, approval no. IRB-2024-941; and center 3: Changzhou No. 2 People’s Hospital, approval no. [2023] 029-05) and adhered to the Declaration of Helsinki. Written informed consent was obtained from all participants. Hangzhou Weiyi Technology provided the robotic equipment, biopsy needles, and insurance coverage for all participants involved in the study. The authors, who are not employees or consultants of the supporting organization, retained complete autonomy over the inclusion of all the data and information submitted for publication.

The inclusion criteria were (a) age 18 to 80 years, (b) clinically suspected lung, liver, or kidney neoplasms, (c) Eastern Cooperative Oncology Group performance status ≤ 1, and (d) willingness to provide informed consent. The exclusion criteria were (a) uncorrectable coagulation dysfunction or anticoagulant therapy within 1 week before the biopsy procedure, (b) body mass index (BMI), calculated as weight in kilograms divided by height in meters squared, greater than 28 (because greater values can affect clearance within the robotic gantry), (c) severe cardiopulmonary, hepatic, or renal insufficiency, and (d) pregnancy or lactation. A stratified, block randomization design (block size = 4) was implemented by independent statisticians using R software (version 4.3.0; R Foundation), with study center as the stratification factor. Participants were allocated in a 1:1 ratio to either robot-assisted or freehand needle insertion.

Robotic System

The robotic system comprises a trajectory-planning workstation and a table-mounted robotic arm with six degrees of freedom; the arm can be detached manually in case of an emergency. After the participant is positioned appropriately, an initial CT scan is performed, and the images are transferred to the system for semiautomatic registration, establishing spatial correspondence between the robot, the participant, and the images. The proceduralist confirms the puncture trajectory and checkpoints. Defined by the operator, checkpoints are reference coordinates along the planned needle trajectory for stepwise insertion verification. Using preprocedural CT images, the proceduralist marks the skin entry point, 1–2-cm–spaced intermediate waypoints, and the lesion target. The robotic system halts at each checkpoint for CT validation of needle tip position. Any excessive deviation triggers recalculation of the remaining trajectory and orientation adjustment. This workflow enables ongoing trajectory correction, reduces cumulative angular errors, and reduces risks of target misplacement and injury to surrounding tissues. The robot then inserts the needle with a controlled force of 11 N, a speed of 2 mm/sec, and a maximum angulation of 25° in the axial and sagittal planes relative to the table normal.

CT-guided Percutaneous Biopsy

All procedures were performed in accordance with the Society of Interventional Radiology guidelines (22). Biopsies were performed using clinical CT scanners (Philips Brilliance 64, Philips Big Bore, or Siemens Perspective) with the following typical settings: 110–120 kVp, 100–200 mA, and a 3-mm section thickness/interval; this was followed by reconstruction to 1-mm sections for planning. All needle insertions were performed or directly supervised by senior proceduralists (H.H., C.C., Y.W., L.X., H.J., and H.Z., with >10 years of experience in image-guided percutaneous interventions). Operators in the robotic group completed standardized phantom training (minimum 5 lesions) before the trial. Proficiency was efficiently achieved via familiarization with the system interface, target and checkpoint definitions, and anesthesia administration.

For freehand procedures, following sterile preparation and local anesthesia (10 mL of 1% lidocaine), the puncture site was determined using the CT gantry laser. The proceduralist inserted the needle using a 17-gauge, 13-cm coaxial introducer via a stepwise approach, adjusting the trajectory with consecutive CT guidance until satisfactory positioning was achieved—namely, maneuvering the needle as close to the preplanned target as possible. A needle tip deviation of less than 4 mm from the target was deemed optimal; if the deviation exceeded 4 mm but the operator determined that the current position could still yield a sufficiently diagnostic sample and that further adjustments could cause complications (eg, hemorrhage, pneumothorax, seeding metastasis) or require multiple adjustments to reduce the deviation, then no further needle adjustment were performed.

For the robot-assisted procedures, the robot was prepositioned on the CT table (setup time ≈ 5 minutes). The robot remained static during the procedure and did not move with the participant, who maintained free breathing throughout the procedure. An initial scan was performed for registration and trajectory planning. After laser-guided skin marking and local anesthesia administration, the robotic arm performed a stepwise insertion technique, by which it advanced the introducer needle incrementally (typically 1–2 cm) toward predefined checkpoints. The position of the needle was verified on interim CT scans, and the trajectory was adjusted as needed. If a deviation was detected, the robotic system recalculated the trajectory through the planning software on the basis of the updated images, and the robotic arm repositioned the needle guide accordingly. Manual adjustment independent of the robot was not permitted by the study protocol. Upon satisfactory positioning, the robotic arm was retracted and the target was manually biopsied.

In both groups, specimens were obtained using an 18-gauge semiautomatic biopsy needle (2-cm throw), with at least two tissue cores collected per lesion, ensuring acquisition of sufficient tissue for histopathology and molecular analysis. The samples were fixed in 10% formalin. Hemostasis was achieved via compression, and a final CT scan was performed to assess the participant for immediate complications.

Outcome Measures and Follow-up

The primary outcome was the technical success rate, defined as the proportion of procedures in which the three-dimensional Euclidean distance (d) from the needle tip (x₁, y₁, z₁) to the physician-preset point (x₂, y₂, z₂) in the target lesion was <4 mm. This threshold was selected on the basis of geometric constraints related to minimum targetable lesion sizes (≥10-mm diameter [ie, 5-mm radius]), established precedents in the robotic intervention literature where mean accuracies of 1–4 mm are consistently reported and deviations > 4 mm are considered outliers (11,21,23), and the requirement for sub–4-mm accuracy to achieve adequate ablation margins (≥5 mm) in therapeutic applications. Specifically, the objective geometric measure was calculated as follows:

$$d=\sqrt{{\left(x_1-x_2\right)}^2+{\left(y_1-y_2\right)}^2+{\left(z_1-z_2\right)}^2}$$

Secondary end points included the following: needle placement deviation or targeting error (Euclidean distance [d]), measured on the prebiopsy confirmation CT images acquired after target positioning of the puncture needle and release of the needle holder; puncture time; number of CT acquisitions and needle adjustments during puncture; safety; and pain.

Puncture time was defined as the interval from cutaneous insertion of the anesthetic needle to final placement of the puncture needle at the predefined target, excluding the time for device installation. Because the robotic device setup depended on the operating room conditions, staffing, and device connections, the full configuration was required only for the initial procedure. The device was therefore premounted on the CT table, and all the data cables were connected before the robotic puncture procedure. Afterward, the participants underwent direct puncture with no repeated preparation steps.

Safety was assessed via vital signs, Eastern Cooperative Oncology Group status, laboratory tests, and adverse events graded per Clavien-Dindo version 2.0 (19,24). Pain was quantified using a visual analog scale (0–10 points; 0 = no pain, 10 = worst imaginable pain) every 2 hours within 24 hours postoperatively. A score ≥ 4 points and/or the need for pharmacologic intervention (including acetaminophen, nonsteroidal anti-inflammatory drugs, or opioids) resulted in the pain being recorded as a complication.

Participants were hospitalized overnight for observation and returned on day 7 for follow-up as a protocol-mandated safety measure to allow rigorous monitoring for any immediate or delayed adverse events. All timing and frequency data were recorded by an independent researcher. Two radiologists (X.L. and H.L., with 9 and 7 years of experience in thoracic and abdominal interventional radiology, respectively), who were masked to the group allocation (robot-assisted vs freehand) and all clinical information of the participants, independently reviewed the postprocedure CT images to assess needle placement deviation. Disagreements were resolved by a senior expert specializing in interventional radiology (Y.G., with >20 years of experience). Data and images were anonymized before evaluation.

Prespecified analyses included the primary end point (technical success rate), all key secondary end points (needle placement deviation, number of CT scans, number of needle adjustments, puncture time, and complication rates), and subgroup analyses by study center and organ, whereas the sensitivity analysis for puncture time excluding the extreme outlier (51.4 minutes) was conducted post hoc.

Statistical Analysis

The sample size was calculated assuming a 95% technical success rate in both groups to test for noninferiority. With a noninferiority margin of −10%, a one-sided significance level (α) of .025, and statistical power of 80%, the required sample size was determined to be 75 participants per group. Accounting for a 10% dropout rate, 168 participants (84 per group) were planned. The intention-to-treat population included all randomized participants with available outcome data. Continuous variables are presented as the means ± SDs and were compared using the t test or the Wilcoxon rank sum test according to the normality of the distribution as assessed with the Shapiro-Wilk test. Categorical variables are summarized as numbers with percentages and were analyzed using the χ² test or the Fisher exact test, as appropriate. Data were analyzed per protocol. Interobserver agreement was evaluated using the intraclass correlation coefficient. All analyses were performed with SAS version 9.4 (SAS Institute) using a two-sided significance level of α = .05. To account for the multicenter design and potential center effects, the Cochran-Mantel-Haenszel test, stratified by study center, was performed as the statistical method for the primary outcome to estimate the treatment difference. In accordance with the International Council for Harmonization E9 statistical principles and a closed testing procedure, if the lower bound of the 95% CI for the difference in success rates was greater than the noninferiority margin (−10%), then noninferiority was established. If the lower bound of the same 95% CI was subsequently greater than 0, then superiority was established without inflating the type I error rate.

Results

Participant and Tumor Characteristics

A total of 165 participants were enrolled across three academic centers from July 2023 to April 2024 and randomly assigned to undergo either robot-assisted (n = 82) or freehand (n = 83) CT-guided percutaneous biopsy. Following randomization, three participants were excluded from the final analysis. Two participants allocated to the robot-assisted group were excluded, one owing to anxiety regarding the robotic device and another owing to an inability to maintain the required immobilization during the procedure. One participant in the freehand group was excluded because of unexpected CT scanner malfunction, which necessitated conversion to US-guided biopsy. These exclusions are detailed in the participant flow diagram (Fig 1). The mean age of the final enrolled participants was 60 years ± 10 (SD), comprising 83 men (50.30% [83 of 165]) and 82 women (49.70% [82 of 165]). There was no evidence of a difference in the baseline demographic and clinical characteristics between the two intervention groups (Table 1). Specifically, the robot-assisted group and the freehand group were comparable in terms of sex distribution (48.78% [40 of 82] male vs 51.81% [43 of 83] male, respectively; P = .70), mean age (59 years ± 10 vs 61 years ± 10, respectively; P = .17), and anthropometric measures, including height, weight, and BMI (P = .71, P = .27, and P = .38, respectively). A single minor protocol deviation was noted: One participant in the robotic group had a BMI of 34.2, which exceeded the prespecified exclusion threshold of BMI > 28. Pre-enrollment verification confirmed full compatibility with the robotic platform, and the participant was retained in all analyses per the intention-to-treat principle. Similarities were also observed in lesion characteristics: The mean lesion size was 26.7 mm ± 15.4 in the robot-assisted group versus 28.0 mm ± 18.0 in the conventional group (P = .61). The 165 lesions were distributed across the lung (n = 103 [62.42%]), liver (n = 50 [30.30%]), and kidney (n = 12 [7.27%]); similar to the other lesion characteristics, there was no evidence of a difference in organ distribution between the groups (P = .46).

Figure 1:

Participant recruitment and study design flowchart. BMI = body mass index, PP = per protocol.

Table 1:

Baseline Characteristics of the Participants and Lesions

Characteristic	Robot (n = 82)	Freehand (n = 83)	P Value
Participants
Sex			.70*
Male	40 (48.78)	43 (51.81)
Female	42 (51.22)	40 (48.19)
Age (y)	59 ± 10 (53–68)	61 ± 10 (55.5–68)	.17†
Height (cm)	163.7 ± 7.5 (158–170)	164.2 ± 7.7 (159.5–170)	.71†
Weight (kg)	60.4 ± 9.7 (54–65)	62.1 ± 10.2 (55–70)	.27†
BMI	22.6 ± 3.4 (19.9–24.9)	23.0 ± 2.9 (21.2–24.9)	.38†
Target lesions			.46‡
Lung	51 (62.19)	52 (62.65)
Liver	28 (34.15)	22 (26.51)
Kidney	3 (3.66)	9 (10.84)
Size (mm)	26.7 ± 15.4 (16.4–30.7)	28.0 ± 18.0 (14.6–38.05)	.61†

Note.—Data are the numbers of participants or lesions, with percentages in parentheses, or means ± SDs, with IQRs in parentheses. BMI = body mass index (calculated as weight in kilograms divided by height in meters squared).

^*χ² test.

^† t test.

^‡ Fisher exact test.

Procedure Outcomes

All the robot-assisted procedures were successfully completed without any device malfunctions or technical failures. Conclusive pathology diagnoses were obtained from all biopsy specimens, indicating a 100% diagnostic success rate for both methods (robotic: 82 of 82, freehand: 83 of 83). The operational workflow of and representative biopsy procedures performed with the robotic system are illustrated in Figures 2 and 3.

Figure 2:

Workflow diagram of the Micro IGT-RoboC (Abacus Surgical) system for needle insertion.

Figure 3:

Clinical photographs of the Micro IGT-RoboC (Abacus Surgical) system and CT-guided percutaneous biopsy in representative cases. (A) The robotic system performing a liver biopsy in a 71-year-old male participant. The planes used to calculate three-dimensional Euclidian distance are noted by the arrows and respective letters z, y, and x. (B) Close-up view of the robotic arm and needle holder gripping the coaxial introducer. (C, D) Robotic needle positioning for biopsy of a lung nodule in a 55-year-old female individual with a history of breast cancer. (E, F) Robot-assisted biopsy of a left liver tumor in a 67-year-old male individual with rectal cancer; the images show introducer needle advancement along the automated trajectory defined by checkpoints (blue and green dots) and position verification. (G, H) Robotic insertion of the coaxial cannula and subsequent verification during biopsy of a left renal mass in a 69-year-old male individual. In C–H, arrows indicate target lesions, and arrowheads indicate the needle holders.

Quantitative comparisons of the procedural outcomes revealed advantages for robotic assistance across multiple metrics (Table 2). The primary end point of technical success was greater in the robot-assisted group (97.56% [80 of 82]; 95% CI: 91.31, 99.40) than in the freehand group (62.65% [52 of 83]; 95% CI: 52.03, 72.23; P < .001). The Cochran-Mantel-Haenszel–adjusted difference in the effective rate was 32.68% (95% CI: 21.08, 44.28); the lower limit of this interval (21.08%) was well above the noninferiority margin of −10%, thus rejecting the null hypothesis and validating the noninferiority hypothesis of the trial. This superior performance in the robot-assisted group versus freehand was consistently maintained across all participating centers (center 1: 100% [41 of 41] vs 73.81% [31 of 42], center 2: 96.43% [27 of 28] vs 53.57% [15 of 28], and center 3: 92.31% [12 of 13] vs 46.15% [6 of 13]) and different lesion locations (lung: 98.04% [50 of 51] vs 57.69% [30 of 52], liver: 96.43% [27 of 28] vs 63.64% [14 of 22], and kidney: 100% [three of three] vs 88.9% [eight of nine]).

Table 2:

Outcomes of Robotic versus Freehand Insertion

Variable	Robot (n = 82)	Freehand (n = 83)	P Value
Successful insertion	80/82 (97.56)	52/83 (62.65)	<.001*
Center 1	41/41 (100.00)	31/42 (73.81)
Center 2	27/28 (96.43)	15/28 (53.57)
Center 3	12/13 (92.31)	6/13 (46.15)
Lung	50/51 (98.04)	30/52 (57.69)
Liver	27/28 (96.43)	14/22 (63.64)
Kidney	3/3 (100.0)	8/9 (88.9)
Targeting error (mm)	1.7 ± 1.1 (1.2–1.9)	4.5 ± 3.9 (2.0–5.2)	<.001†
No. of CT scans	4.3 ± 1.9 (3–5)	5.2 ± 2.3 (4–6)	.002‡
No. of needle adjustments	0.7 ± 0.7 (0–1)	1.6 ± 1.6 (0–2)	.003‡
Puncture time (min)	5.5 ± 4.3 (2.9–7.6)	4.8 ± 7.0 (1.4–5.2)	.50†

Note.—Data are the numbers of participants or lesions, with percentages in parentheses, or means ± SDs, with IQRs in parentheses. Robotic insertion achieved significantly higher technical success rates, lower targeting errors, fewer CT scans, and fewer needle adjustments compared with freehand insertion, with no statistically significant difference in puncture time between the two groups.

^*Fisher exact test.

^† t test.

^‡ Wilcoxon rank sum test.

Needle placement deviation, measured as the Euclidean distance from the needle tip to the preset point in the target lesion, was better with robotic assistance (1.7 mm ± 1.1) than with the freehand technique (4.5 mm ± 3.9, P < .001). The robotic approach also resulted in the need for fewer needle adjustments (0.7 ± 0.7 vs 1.6 ± 1.6, P = .003) and CT scans (4.3 ± 1.9 vs 5.2 ± 2.3, P = .002) for achieving optimal needle positioning.

The interobserver agreement for all the radiologic measurements was excellent, with an intraclass correlation coefficient of 0.94 (95% CI: 0.92, 0.96), indicating near-perfect reliability between the two masked radiologists. The mean targeting error was 3.16 mm ± 3.21 and 3.26 mm ± 3.53 for the two readers, respectively, yielding an overall average of 3.21 mm ± 3.33. Variance decomposition revealed that between-subjects variability accounted for 96.99% of total variance, while interobserver variability contributed only 3.01% (mean square between subjects = 22.14, mean square within subjects = 0.68).

Robot-assisted biopsy achieved procedural efficiency comparable to that of experienced operators, with no evidence of a difference in mean puncture time with the freehand biopsies (5.5 minutes ± 4.3 vs 4.8 minutes ± 7.0, P = .50). Notably, one participant in the freehand group with a Clavien–Dindo grade III complication (pulmonary hemorrhage) required a puncture time of 51.4 minutes, which could have biased the statistical results and affected the statistical power. Therefore, a sensitivity analysis was performed by excluding the outlier to further validate the findings; the results still demonstrated comparable puncture time between the groups (5.5 minutes ± 4.3 vs 4.2 minutes ± 4.2, P = .07), confirming the robustness of the primary analysis.

Adverse Events

During the study period, a total of 42 procedure-related complications were recorded in 28 of the 165 participants (16.97%). The overall incidence of complications was lower in the robot-assisted group (12.20% [10 of 82]) than in the freehand group (21.69% [18 of 83]), although there was no evidence of a difference (P = .10). However, subgroup analysis based on lesion location revealed important differences between the groups (Table 3).

Table 3:

Adverse Events Occurring in Participants

Adverse Event	Robot (n = 82)	Freehand (n = 83)	P Value
Lung	4/82 (4.88)*	14/83 (16.87)*	.014†
Pain (grade I)	2	8
Pneumothorax (grade I)	2	5
Dyspnea (grade I)	0	4
Nausea (grade I)	0	3
Cough (grade I)	0	2
Hydropneumothorax (grade II)	0	2
Low blood pressure (grade I)	0	2
Intra-alveolar hemorrhage (grade I)	0	1
Hemoptysis (grade III)	0	1
Liver	6/82 (7.32)*	4/83 (4.82)*	.50†
Pain (grade I)	5	3
Pneumothorax (grade I)	1	1
Hepatic hemorrhage	0	0
Kidney	0	0
Renal hemorrhage	0	0
Total	10/82 (12.20)*	18/83 (21.69)*	.10‡

Note.—A total of 42 complications were reported by 28 participants: 14 participants in the freehand lung biopsy group accounted for 28 complications, while the remaining 14 participants (from other sites and groups) each reported a single complication. Complications were graded according to the Clavien–Dindo version 2.0 guidelines. Overall adverse event rates were comparable between groups, but robotic lung biopsy was associated with a significantly lower complication rate than freehand lung biopsy; events were predominantly mild (Clavien-Dindo grade I).

^*Data represent the numbers of participants with adverse events, with percentages in parentheses.

^† Fisher exact test.

^‡ χ² test.

For lung biopsies, the robot-assisted approach resulted in a lower complication rate (4.88% [four of 82]) compared with the freehand technique (16.87% [14 of 83], P = .014). The pulmonary complications experienced in the freehand group included pain (n = 8), pneumothorax (n = 5), dyspnea (n = 4), nausea (n = 3), cough (n = 2), hydropneumothorax (n = 2), low blood pressure (n = 2), intra-alveolar hemorrhage (n = 1), and hemoptysis requiring intervention (n = 1). In contrast, the participants in the robot-assisted group experienced only four complication events—two pain events and two pneumothorax events—all of which were classified as Clavien-Dindo grade I. One serious adverse event (pulmonary hemorrhage with Clavien-Dindo grade III hemoptysis) occurred in the freehand group; this event was managed successfully with endoscopic clot evacuation and was assessed by the independent safety committee as possibly related to vessel injury during the manual needle insertion process.

With respect to liver biopsies, the complication rates were comparable between groups (robot-assisted: 7.32% [six of 82] vs freehand: 4.82% [four of 83], P = .50), with pain being the most frequently reported adverse event in both groups. No biopsy-related complications occurred in either group for kidney procedures. All other complications were mild (grade I) or moderate (grade II) in severity and resolved with appropriate conservative management.

Discussion

Surgical robotic systems have become powerful tools in CT-guided percutaneous interventions (2,5,7,19,21). The results of our prospective multicenter randomized trial demonstrate that compared with the conventional freehand technique, a novel robotic system improves the precision and safety of percutaneous biopsy for thoracoabdominal lesions. Given the clinical challenges associated with differences in operator experience across different centers and the strict definition of technical success (needle tip to the physician-preset point in the target lesion < 4 mm), we evaluated a robotic system designed to automate trajectory planning and needle insertion. Our findings confirm that robotic assistance yields higher technical success rates (97.56% vs 62.65%, P < .001) and needle placement deviation (1.7 mm ± 1.1 vs 4.5 mm ± 3.9, P < .001) compared with freehand technique and requires fewer needle adjustments and CT scans. The reported 62.65% technical success rate in the freehand group is lower than that in the established literature (25). This discrepancy is entirely attributable to our strict <4 mm geometric threshold for technical success. The surgeons in the freehand group successfully obtained adequate samples even when the needle tip deviated by >4 mm, particularly from larger lesions. The ability of the robotic system to consistently achieve <4-mm targeting error highlights its mechanical precision, which may be particularly valuable for sampling very small lesions or establishing complex trajectories where manual adjustments are challenging.

The comparable mean puncture times between robot-assisted and freehand biopsy (5.5 minutes ± 4.3 vs 4.8 minutes ± 7.0, respectively; P = .50) indicate that the robotic system did not improve procedural efficiency. The wide range in freehand puncture times, however, underscores the inconsistency of the manual approach, whereas the robotic system provided a more standardized workflow. Even after excluding outliers, sensitivity analysis confirmed comparable puncture times between the two groups (5.5 minutes ± 4.3 vs 4.2 minutes ± 4.2, P = .07), confirming the robustness of the primary analysis. However, given the similar clinical outcomes (100% diagnostic yield [82 of 82 and 83 of 83]) and procedural times, the high capital cost of the robotic system remains a consideration. Formal cost-effectiveness analyses are needed to determine the value proposition of robotic assistance, which may lie in standardizing precision for less-experienced operators and reducing complications in high-risk participants.

Our results align with and extend the growing evidence supporting the use of robotic assistance in percutaneous interventions. The targeting error of 1.7 mm achieved in our study is consistent with that of high-performance systems such as the Xact system (1.6 mm) (21) and the AcuBot (1.2 mm) (20). In a separate first-in-human trial, the floor-mounted Zerobot system enabled CT fluoroscopy–guided introducer needle placement within a predefined ≤10-mm needle tip to lesion margin in 10 participants, with a mean puncture time of 4 minutes (range, 0–23 minutes) (19). A randomized study of 14 liver ablations revealed reductions in the mean number of needle adjustments (one vs 3.7 ± 1.3), mean procedure time (3.6 minutes ± 1.1 vs 8.6 minutes ± 2.0), and mean radiation exposure for patients (469.7 mrem ± 177.1 vs 7075.7 mrem ± 3181.7) with robotic insertion versus manual insertion (all P < .05); however, that system is yet to enter the market (26). In a recent multicenter randomized trial involving 140 participants with thoracoabdominal tumors, a CT-guided robotic positioning system improved technical success (84.06% [58 of 69] vs 43.66% [31 of 71]), reduced the number of needle adjustments (1.5 ± 1.5 vs 3.5 ± 3.1), and decreased the number of required CT scans (5.0 ± 2.1 vs 7.1 ± 3.7); however, there was no evidence of a difference in procedure time or complication rates (27). Notably, the optical tracking component of the system required an unobstructed path because physical interference could compromise marker recognition.

Radiation exposure remains a concern in CT-guided biopsy. Conventional manual techniques frequently require repeated needle repositioning and associated confirmatory imaging, increasing cumulative radiation doses (28). A systematic review estimated that robotic assistance can reduce radiation exposure by approximately 50% while maintaining targeting accuracy (29). A recent retrospective study that compared the efficacy of percutaneous lung biopsy using a patient-mounted Ace robotic system (XACT Robotics) with that of manual needle insertion under CT fluoroscopy guidance revealed comparable technical success rates (86.67% [13 of 15] vs 89.39% [59 of 66], respectively; P = .67) and a 100% diagnostic yield in the robotic group (15 of 15) (30). Unlike manual CT fluoroscopy, which exposes the operator to unavoidable in-room radiation, robotic systems allow remote control and completely eliminate such occupational exposure. In our study, the robotic system required fewer CT scans than the freehand technique did (4.3 ± 1.9 vs 5.2 ± 2.3, P = .002), indicating a corresponding reduction in patient radiation exposure.

All reported adverse events were procedure-related complications commonly associated with percutaneous biopsy. Among lung biopsies specifically, the robot-assisted group experienced fewer complications than the manual technique group did (P = .014). Previous studies have reported that a prolonged procedure time and multiple pleural transgressions are independent risk factors for pneumothorax and hemorrhage (31,32). Despite the time required for image transfer, the total puncture time was comparable between the groups but was more consistent in the robot-assisted group, in which all the insertions were completed within 10 minutes. The robot-assisted approach required fewer needle adjustments than the freehand technique did (0.7 ± 0.7 vs 1.6 ± 1.6, P = .003), providing a mechanistic explanation for the reduced complication rate. The robotic system demonstrated high operational safety throughout the study, with no device malfunctions or equipment-related adverse events reported, supporting its clinical application for percutaneous biopsy.

Rather than evaluating robotic biopsy platforms merely based on accuracy superiority, their clinical merit is defined by noninferior diagnostic performance alongside multifaceted practical merits covering radiation safety, procedural standardization, and long-distance clinical scalability. Mounted robotic manipulators operated via shielded external consoles eliminate intraoperative occupational radiation exposure for operators and concurrently reduce patient radiation burden by cutting redundant CT scanning relative to conventional manual puncture. Benefiting from semiautomated path planning and rigid mechanical positioning, robotic systems homogenize thoracoabdominal percutaneous biopsy workflows and weaken interoperator heterogeneity stemming from disparate clinical experience; decreased repetitive needle tuning further permits junior practitioners to attain consistent expert-level puncture accuracy under physician oversight and streamline hospital procedural management. Additionally, inherent mechanical steadiness and remote manipulation functionality pave the way for teleoperated interventional procedures, facilitating remote expert-guided biopsy deployment in resource-limited rural regions deficient in on-site interventional radiologists and broadening population access to standardized oncologic interventional services.

Several limitations of this study should be acknowledged. First, although the sample size is among the largest in robotic intervention trials to date, it remains insufficient for robust subgroup analyses, particularly for kidney lesions. Second, direct radiation dose measurements were not obtained, despite the observed reduction in the number of CT scans performed in the robotic group. Third, pathology outcomes were not systematically recorded, as diagnostic yield can be influenced by multiple factors beyond needle positioning accuracy. Fourth, the exclusion of participants with BMI > 28 limits the generalizability of our findings to individuals with obesity, despite the successful enrollment of one participant with BMI of 34.2 after confirming full robotic platform compatibility and unimpeded procedural performance. A wider range of robotic models will become available for clinical selection in future studies. Fifth, key determinants of procedural complexity, such as the needle trajectory angle (in-plane vs out-of-plane), skin to target distance, proximity to major blood vessels or critical structures, and the presence of underlying lung parenchymal disease (eg, emphysema or fibrosis), were not systematically recorded in our database. Although the randomized, multicenter design of this trial likely distributed these unmeasured confounding variables evenly between the two groups, the lack of these detailed case-mix characteristics limits our ability to perform stratified analyses based on procedural difficulty. Finally, the system’s respiratory gating function was not activated in this study; instead, we used a stepwise insertion technique to compensate for respiratory motion.

In conclusion, in this multicenter randomized trial, the CT-guided robotic system demonstrated safety and accuracy for percutaneous biopsy of thoracoabdominal lesions. Compared with the freehand approach, robotic assistance achieved superior geometric targeting precision, reduced procedural variability, and decreased complication rates associated with lung biopsies. However, both methods achieved excellent diagnostic yields, and the robotic system did not reduce the total procedural time. The primary clinical value proposition of this robotic system lies not merely in its geometric accuracy, but in its potential to dramatically reduce operator radiation exposure, standardize procedural quality across varying levels of operator experience, and facilitate future teleoperated interventions. Future studies should validate these findings in larger populations, incorporate direct radiation dosimetry, assess cost-effectiveness, compare robotic assistance with contemporary optimized manual techniques, and explore the integration of multimodality imaging and advanced motion compensation to further expand the capabilities of robotic-assisted intervention.

Disclosures of conflicts of interest

Please see ICMJE form(s) for author conflicts of interest. These have been provided as supplemental materials.