This is a nonrandomized prospective study with approval from the Institutional Review Board and consent from all the patients. The calculated sample size of cases needed in each group was 96 before and after cooperative training with an 80% power value and a 5% α value (the type I error probability for a two sided test), according to the PS Power and Sample Size Calculations (version 2.1.30, Noshville, TN).

Between September 2003 and August 2004, 200 nodular HCCs in 146 consecutive patients were prospectively recruited to undergo planning US before their radiofrequency ablation procedures. We only included the patients who were recommended to undergo US guided percutaneous radiofrequency ablation of HCC after a triage by five hepatologists who had at least 3-years of experience in referring the patients for US-guided radiofrequency ablation. Of the 146 patients, 140 (96%) had risk factors of HCC (liver cirrhosis: 86% and chronic hepatitis: 10%). The levels of α-fetoprotein (AFP) were highly variable with a range of 4-12,794 ng/mL (mean: 15 ng/mL). We divided the patients into two groups: Group I included 74 patients with 100 nodular HCCs which were evaluated between September 2003 and February 2004 before the cooperative training. Group II was composed of 72 patients with 100 HCCs who were evaluated between March 2004 and August 2004 after the cooperative training. The amount of time between the before and after the cooperative training was one month. The clinical characteristics of the patients in each group are summarized in Table 1. All patients had HCCs detected at contrast-enhanced multiphase helical CT. The diagnosis of HCC for 13 tumors was assessed by percutaneous needle biopsy. The remaining 187 tumors were considered to be HCCs on the basis of imaging findings (n = 157, characteristic enhancement patterns at contrast-enhanced multiphasic CT or dynamic contrast-enhanced MR imaging - hypervascularization during the hepatic arterial phase, and a wash-out pattern during the equilibrium phase) and elevated serum tumor marker (n = 5, α-fetoprotein [AFP] level > 400 ng/mL [> 400 µg/L]) or at least two concurrent radiological findings which were compatible with HCC among CT, MR imaging, US, and angiography (n = 25) (20).

All patients met the following criteria for treatment with percutaneous radiofrequency ablation: the presence of a single tumor with a maximum diameter not exceeding 5 cm, or the presence of multiple (not more than four) tumors and each with a maximum diameter not greater exceeding 3 cm; the absence of portal venous thrombosis and extrahepatic metastasis; the presence of liver cirrhosis of Child-Pugh class A or B. Remote recurrent HCCs (n = 88) after hepatectomy (n = 13), percutaneous ethanol injection therapy (n = 2), transarterial chemoembolization (TACE) (n = 13), radiofrequency ablation (n = 39) and combined treatment (n = 21) were also included in the study irrespective of the prior history of treatment. The operability of the patients did not affect the inclusion criteria.

We used 17-gauge internally cooled single or cluster RF electrodes (Radionics, Burlington, MA) with a 500-kHz monopolar generator (series CC-3; Radionics) which was able to produce a 200-W output.

Three abdominal radiologists, who had different levels of experience for the diagnosis of percutaneous radiofrequency ablation of hepatic tumors, independently performed the sonographic examinations for each tumor by themselves. Immediately after one observer finished the sonographic examination for each patient, the others followed within a mean of 5 minutes. They were aware of the CT findings before conducting the sonographic studies. Observer 1 (senior attending staff) assessed more than 1,000 cases of percutaneous radiofrequency ablation under the guidance of US, observer 2 (junior staff) assessed over 150 cases and observer 3 (fellow) had 15 cases of experience without doing the procedure during the research time. All the sonographic examinations were performed using 1.0- to 4.0-MHz convex array probes (Acuson-Sequoia 512, Siemens Medical Solutions, Mountain View, CA) or 2.0- to 5.0-MHz curvilinear probes (HDI 5000; Advanced Technology Laboratories, Philips Medical Systems, Bothell, WA). Using gray-scale and color Doppler US, the observers independently evaluated the feasibility of US guided percutaneous radiofrequency ablation of each tumor. They also archived the sonographic images of each tumor on a picture archiving and communication system (PACS, Centricity Workstation, version 1.0; GE Medical Systems, Milwaukee, WI). The three radiologists scanned 74 patients with 100 nodular HCCs before the cooperative training and they decided the feasibility of the procedure for each patient based on their experience. After the cooperative training, the remaining 72 patients with 100 HCCs were independently examined by the same three radiologists.

After the sonographic examination, they recorded the results for the feasibility of performing US guided percutaneous radiofrequency ablation for each tumor. The probability of completeness of treatment (i.e. incomplete ablation or primary effectiveness) did not affect the feasibility. We simplified the feasibility as "feasible" or "not feasible". In general we considered the following tumors as ineligible for US-guided percutaneous radiofrequency ablation: 1) When the critical structures (large vessels, lung base, gallbladder or gastrointestinal tract) are located in the presumed electrode pathway (i.e. When the gallbladder or gastrointestinal tract interrupts the pathway of electrode, or when we cannot keep away from the segmental portal vein or main portion of hepatic vein or vessel greater than 3 mm in diameter); 2) When the tumor is not visualized on US scans. Whenever the observers considered that percutaneous radiofrequency ablation was not feasible, they were asked to describe the reasons why it was not feasible. This work was done independently by each observer. After the planning US, the final treatment methods of these patients or outcome of radiofrequency ablation were evaluated.

The training sessions were performed after the sonographic studies done for group I (100 nodular HCCs in 74 patients). The modulator (observer 3) arranged the training and prepared all the necessary data only. All three observers participated in the cooperative training for four hours which involved the review of the sonographic images archived on the PACS workstation for all the cases which showed different decisions on the feasibility among the observers. Immediately after reviewing each case, we discussed the reason for the discrepancy and we tried to reach a consensus on the feasibility of performing percutaneous radiofrequency ablation for cases which were not the decision of senior staff. We compared the index tumor on US with the presuming lesion on the CT scan for evaluation of the reason of non-visualized cases. If the tumor had low conspicuity, a radiologist tried to make a higher conspicuity with use of harmonic images and taught it to other radiologists. We evaluated the location of the cases which showed a discrepancy in the feasibility between the operators and analyzed the critical structures which were obstacles to the the safe route to approach the index tumor. We discussed if there were an alternative safe route to approach the tumor. The same process was continued for the all cases with different decisions on feasibility.

The overall feasibility and individual feasibility rates were evaluated before and after the cooperative training. Also, the reasons for all the cases in which percutaneous radiofrequency ablation was considered not feasible were assessed. Moreover, before and after the cooperative training, the inter-observer agreement among the three observers on the feasibility were analyzed. Furthermore, for the cases with a discrepancy between the three observers, the characteristics of the tumors were assessed and the causes of these cases were evaluated. The difference in the numbers showing a discrepancy were assessed before and after cooperative training.

The feasibility rates of the observers between the two groups were statistically analyzed using the Generalized Estimating Equation (GEE) test. To assess the agreement between the observers on the feasibility of performing US guided percutaneous radiofrequency ablation, kappa (κ) values were calculated using a computer program (MedCalc, version 6.0; Medcalc Software, Mariakerke, Belgium). A κ-value greater than zero was considered to be indicative of a positive correlation. The inter-observer agreement was rated as follows: poor, κ<0.20; fair, κ = 0.21-0.40; moderate, κ = 0.41-0.60; good, κ = 0.61-0.80; very good, κ = 0.81-1.00. This was assessed by comparing the Cohen's κ coefficients (21, 22) which were calculated for each pair of observers before and after the cooperative training. Comparisons between the κ values before and after the cooperative training were performed using the Z test.

The numerical formula for the Z test is:

(K1; κ value of group I, K2; κ value of group II SE1; standard error of group I, SE2; standard error of group II)

A p-value less than 0.05 was considered to indicate a statistically significant difference.

The overall feasibility rate judged by the proportion of the tumors considered eligible for US guided percutaneous radiofrequency ablation by all three observers (Figs. 1, 2), was 73% (436/600). The feasibility rates of each observer for group I were 71% (observer 1; 71/100), 77% (observer 2; 77/100) and 70% (observer 3; 70/100) whereas, for group II were 73% (observer 1; 73/100), 76% (observer 2; 76/100) and 69% (observer 3; 69/100), respectively. There was no significant difference in the feasibility rates between the two groups (GEE test, p>0.05).

For all the tumors (n = 164) considered ineligible for US guided percutaneous radiofrequency ablation by all three observers, the most common causes for refraining from performing a radiofrequency ablation procedure included nonvisualization of the tumor (62%, 102/164) and the absence of a safe route for the percutaneous approach (38%, 62/164) (Fig. 3).

The mean tumor size considered ineligible for US guided percutaneous radiofrequency ablation was 1.9 cm (range, 0.7-3.4 cm), whereas the mean tumor size considered eligible was 2.1 cm (range, 0.5-5.0 cm). The difference in size between the two groups was not statistically significant (p > 0.05, student T-test). The reasons for the cases which had no safe treatment route (n = 62) included abuttment on the large hepatic vessels (n = 23), obscured by the lung (n = 19), abuttment on the gallbladder (n = 6), abuttment on the heart (n = 4), exophytic growth (n = 4), abuttment on the diaphragm (n = 2), attachment to the bowel (n = 2), and attachment to the inferior vena cava (n = 2).

After the planning US, 171 patients were treated and among these, 49% (84/171) underwent US guided percutaneous radiofrequency ablation. After radiofrequency ablation, only 5% (4/84) of patients was a residual tumor found and those four patients were treated by TACE for the residual tumor. The rest of the 87 patients were treated with TACE (n = 66), intra-operative radiofrequency ablation (n = 7), surgical resection (n = 9), and liver transplantation (n = 4) (Table 2).

We found moderate (κ-value, 0.41-0.60) inter-observer agreement for all three observers in group I and good inter-observer agreement (κ-value = 0.575) for group II (Table 2). The κ-values between the observers for group I were 0.575 (observers 1 and 2), 0.495 (observers 1 and 3), and 0.431 (observers 2 and 3), whereas those for group II had κ-values of 0.763 (observers 1 and 2), 0.758 (observers 1 and 3), and 0.676 (observers 2 and 3), respectively. The improvement of inter-observer agreement between observers 1 and 3, and 2 and 3 was statistically significant (Z-test) (Table 3) before and after the cooperative training (p<0.05). The different kappa values of observers 1 and 2 were not statistically significant between both groups (p>0.05).

We observed a significant difference in the number of lesions n = 32) with different feasibility assessments between the three observers before the cooperative training and 18 lesions after the cooperative training (p<0.05, Chi-square test). The mean diameter of the coincident cases was 2.2 cm (range, 0.5-5.0 cm) whereas tumors which had different feasibility assessments were on average, 1.8 cm (range, 0.7-3.5 cm) (p < 0.05, Mann-Whitney U test). The frequency of tumor locations was 27% in the left hepatic lobe and 73% in the right hepatic lobe for coincident cases, whereas the frequency was 36% in the left lobe and 64% in the right lobe for tumor with different feasibility assessments. Before the cooperative training, the most common cause of discrepancy was the ability of localization of the index tumor (72%) and the second cause was the different decision to approach the tumor when the critical structures abutted the tumor (28%) (Fig. 4). After cooperative training, the ability of localization of the tumor had improved (44%), however, the improvement was not statistically significant (p>0.05, Chi-square).