Histopathological diagnosis of microvascular invasion in hepatocellular carcinoma: Is it reliable?

Abstract

BACKGROUND

Microvascular invasion (MVI) is a critical prognostic factor for postoperative hepatocellular carcinoma recurrence, but the reliability of its current pathological diagnosis remains uncertain.

AIM

To evaluate the accuracy of current 7-point sampling methods and propose an optimal pathological protocol using whole-mount slide imaging (WSI) for better MVI detection.

METHODS

We utilized 40 New Zealand white rabbits to establish VX2 liver tumor models. The entire tumor-containing liver lobe was subsequently obtained, following which five different sampling protocols (A-E) were employed to evaluate the detection rate, accuracy, quantity, and distribution of MVI, with the aim of identifying the optimal sampling method.

RESULTS

VX2 liver tumor models were successfully established in 37 rabbits, with an incidence of MVI of 81.1% (30/37). The detection rates [27% (10/37), 43% (16/37), 62% (23/37), 68% (25/37), and 93% (14/15)] and quantity (15, 36, 107, 125, and 395) of MVI increased significantly from protocols A to E. The distribution of MVI showed fewer MVIs farther away from the tumor, but the percentage of MVI detected quantity gradually increased from 6.7% to 48.3% in the distant nonneoplastic liver tissue from protocols A to E. Protocol C was identified as the optimal sampling method by comparing them in sequence. The sampling protocol of three consecutive interval WSIs at the tumor center (WSI₃) was further screened to determine the optimal number of WSIs. Protocol A (7-point sampling method) exhibited only 46% accuracy and a high false-negative rate of 67%. Notably, the WSI₃ protocol improved the accuracy to 78% and decreased the false-negative rate to 27%.

CONCLUSION

The current 7-point sampling method has a high false-negative rate in MVI detection. In contrast, the WSI₃ protocol provides a practical and effective approach to improve MVI diagnostic accuracy, which is crucial for hepatocellular carcinoma diagnosis and treatment planning.

Key Words: Hepatocellular carcinoma; Microvascular invasion; Histopathological diagnosis; Whole-mount slide imaging; Sampling protocol

Core Tip: This study investigated the effectiveness of various pathological sampling protocols for detecting microvascular invasion (MVI) in hepatocellular carcinoma. Conventional 7-point sampling methods have a high false-negative rate for MVI detection. However, the study revealed that whole-mount slide imaging, particularly when three consecutive interval whole-mount slide imaging from the tumor center are used, significantly improves diagnostic accuracy. This innovative approach improves the detection of MVI, especially in distant nonneoplastic liver tissue that have previously been ignored, offering a more reliable method for postoperative prognosis.

INTRODUCTION

Hepatocellular carcinoma (HCC) is a highly lethal malignancy, ranking as the second most fatal tumor globally and the fourth leading cause of cancer-related death[1,2]. China alone accounts for approximately half of all HCC cases worldwide and has the highest incidence and mortality rates[3]. A mere 18% 5-year survival rate is reported[1], with over 90% of HCC-related deaths attributed to metastasis and recurrence[4]. The postoperative recurrence rate of HCC is alarmingly high, reaching 70%[5,6]. Microvascular invasion (MVI) significantly increases the risk of postoperative recurrence in HCC patients[7-9], increasing the recurrence rate by 4.4 times[10].

MVI, characterized by the presence of cancer cell nests within an endothelial-lined vascular lumen observed through microscopic examination[11], holds paramount importance as a pathological indicator for assessing recurrence risk and determining therapeutic strategies for HCC[6]. Postoperative histology serves as the established reference standard for MVI diagnosis. However, the detection rate of MVI ranges widely from 7.8% to 74.4%[12], and the sampling protocol used can lead to false-negative results. In 2015, the Chinese guidelines introduced the 7-point sampling method[11], which is widely utilized in clinical pathological examinations of MVI in HCC patients. Notably, previous studies have shown higher MVI-positive rates with the 7-point sampling protocol than with the 3-point sampling protocol (47.1% vs 34.5%), but similar results were obtained with the 13-point sampling method (47.1% vs 51.3%)[13]. Another study revealed that the MVI detection rate reached a plateau when the number of sampling sites exceeded a certain threshold[14]. These findings collectively support the feasibility of using the 7-point sampling approach for MVI detection, emphasizing the limited benefits of surpassing the number of sampling sites threshold.

In a recent study by Yu et al[15], whole-mount slide imaging (WSI) outperformed the 7- and 3-point sampling protocols in detecting MVI (63.7% vs 33.0% or 22.0%), indicating limitations in the current utilization of the 7-point sampling method. Furthermore, the reliability of 7-point sampling for detecting MVI has not been effectively verified, and this protocol has not been widely adopted outside China[12]. Thus, further investigations are warranted to determine the accuracy of the 7-point sampling method in clinical practice. In this study, we developed an animal model using rabbit VX2 in situ liver tumors to investigate the presence of MVI. Serial WSIs of the entire liver lobe were obtained to comprehensively assess peritumoral MVI. Our aim was to evaluate the diagnostic accuracy of various sampling protocols for MVI and develop a practical and viable pathological sampling approach that improves MVI diagnostic precision. Our findings have significant implications for diagnosing MVI and will help optimize future pathological examination strategies.

MATERIALS AND METHODS

Animals

The ARRIVE guidelines were followed strictly during the conduct of the animal experiments in this study[16]. The protocol was approved by the Institution Animal Care and Use Committee of Zhuhai BesTest Bio-Tech Co., Ltd. in China, protocol number: No. IAC(L)202104001. Forty New Zealand white rabbits weighing 2-3 kg and aged three to four months were used to construct the animal model. The rabbits were housed at Zhuhai BesTest Bio-Tech Co., Ltd. (license number: SYXK 2020-0229), each in an individual cage under trained staff supervision. The facility had a 12-hour light/dark cycle with daylight simulation lamps and good ventilation and air filtration systems. The environmental conditions, such as temperature (20-25 °C) and humidity (60%-70%), were well controlled. Cages and waste trays were cleaned daily to maintain a clean and quiet environment (noise ≤ 60 dB). Food and water were provided freely. Any distress was promptly addressed, and painkillers or anesthetics were administered as needed. The sample size of 40 was established through a statistical power analysis, drawing upon previously published detection rates for various sampling techniques[15], with a two-tailed significance level of 0.05 and a power of 0.8. Factoring in an anticipated 10% sample attrition, this calculation resulted in a target sample size deemed sufficient to ensure statistical robustness.

Anesthesia was induced in the rabbits via intramuscular injections into their hind legs. We used a combination of 0.1 mL/kg xylazine hydrochloride (Dunhua Shengda Animal Medicine Co., Ltd., China), 0.1 mL/kg tiletamine hydrochloride and zolazepam hydrochloride (VIRBAC, Inc., France) for the injection. In accordance with the literature[17,18], we implanted a single VX2 tumor fragment (approximately 1 mm³) provided by Zhuhai BesTest Bio-Tech Co., Ltd. into the left lobe parenchyma of the liver under ultrasonographic guidance via a 16G puncture needle. Tumor growth was monitored continuously by ultrasound until the tumor reached a size of 1-3 cm, indicating successful establishment of the HCC animal model. The weight of the rabbits, along with the tumor growth time and size, were recorded periodically.

Specimen processing

Upon successful establishment of the rabbit VX2 liver tumor model, all the animals were euthanized with air embolism under double-dose anesthesia. The entire liver lobe containing the tumor was then fixed in 4% paraformaldehyde for more than 48 hours. The specimens were subsequently sectioned along the maximal tumor section into consecutive 3-mm-thick tissue blocks parallel to the maximal tumor section, as illustrated in Figure 1A. Each tissue block was placed in the same orientation and then assigned a label with consecutive numbers (as shown in Figure 1B) to maintain accurate specimen identification. If a tissue block was excessively large, it was divided into two parts to ensure appropriate paraffin embedding.

Figure 1 Schematic flowchart illustrating the process of producing whole-mount slide images. A: The entire liver lobe containing the tumor was sectioned in parallel at a uniform thickness of 3 mm along the maximal tumor section; B: Tissue blocks were consecutively numbered and oriented in the same direction for embedding; C: The first complete whole-mount slide images stained with hematoxylin and eosin was prepared for each tissue block. Whole-mount slide images labeled as number 4 correspond to the maximal tumor section, whereas whole-mount slide images labeled as numbers 3-5 correspond to sections containing the tumor at 3 mm intervals. Whole-mount slide images labeled as numbers 1-7 correspond to sections of the entire liver at 3 mm intervals.

WSIs for histological sections

The tissue blocks were dehydrated using an automated tissue dehydrator, embedded in paraffin in the same marked orientation, and then shaped using a paraffin embedding station and mold. Only the first complete tissue section, with a thickness of 5 μm, was prepared from each wax block. These tissue sections were stained with hematoxylin and eosin and digitized into WSIs. For this digitization process, a Pannoramic P250 Flash slide scanner (3D Histech, Budapest, Hungary) was used with the following specific settings: We employed a 20 × objective lens (a resolution of 0.5 μm per pixel), and the output resolution was set to 41 × (0.25 μm per pixel). The scanner was configured for automatic focusing and utilized a bright-field illumination source. With respect to the sample scanning threshold, an automatic setting was employed, and we opted to scan the entire slide (as shown in Figure 1C). The digital images were viewed and analyzed via CaseViewer 2.4 software.

Sampling protocol

The sampling strategies (protocols A-E) implemented in our study are illustrated in Figure 2.

Figure 2 Schematic illustration of various sampling protocols based on whole-mount slide images. A: Protocol A, six regions (labeled a-f) in the maximal tumor section; B: Protocol B, a single complete whole-mount slide image of the maximal tumor section (number 4); C: Protocol C, whole-mount slide images that included the tumor sections at 3 mm intervals (numbers 3-5); D: Protocol D, whole-mount slide images that included the entire liver at 3 mm intervals (numbers 1-7); E: Protocol E, whole-mount slide images that included the entire liver at 0.3 mm intervals.

Protocol A (7-point sampling): Based on the 7-point sampling protocol[11], we preoutlined six specific regions on the WSI of the maximal tumor section (Figure 2A) via CaseViewer 2.4 software while excluding the intratumoral MVI region because it lacks prognostic value[19]. These regions were located at the 12:00, 3:00, 6:00 and 9:00 positions, as were the tumor-adjacent and distant non-neoplastic regions, which are labeled a-f in Figure 2A in sequence. The selected regions were subsequently subjected to examination for the diagnosis of MVI.

Protocol B (WSI sampling): The sampling range of this protocol was the entire WSI of the maximal tumor section (Figure 2B). This approach provided a comprehensive view of the maximal tumor section without focusing on specific subregions, as in protocol A.

Protocol C (tumor 3 mm interval continuous WSI sampling): This protocol’s sampling range was an expansion of that in protocol B. It not only cover the WSI of the maximal tumor section, but also included the adjacent tumor section WSIs. In other words, it included all WSIs of the tumor at 3 mm intervals (Figure 2C) to permit a more extensive examination of the tumor and its adjacent areas.

Protocol D (liver 3 mm intervals continuous WSI sampling): The sampling range was all WSIs of the entire liver lobe, equivalent to obtaining all WSIs of the entire liver lobe at 3 mm intervals (Figure 2D). This provided a broader perspective of the entire liver lobe compared with the previous protocols.

Protocol E (liver 0.3 mm interval continuous WSI sampling): In this protocol, fifteen rabbit liver tumor models were randomly selected. All the tissue blocks of the entire liver lobe were sectioned into serial WSIs at 0.3 mm intervals for MVI examination. The sampling range was equivalent to obtaining all WSIs of the entire liver lobe at 0.3 mm intervals (Figure 2E), which enabled a detailed examination of the entire liver lobe with a relatively fine sampling interval.

Pathologic analysis

In accordance with the guidelines for the diagnosis and treatment of primary liver cancer in China, satellite nodules are considered to be in the progressive stage of MVI during the intrahepatic metastasis of HCC. In this study, satellite nodules ≤ 1 mm that were difficult to differentiate from MVI were included in the MVI category[6]. All the sections were examined independently for MVI by two pathologists with over eight years of experience in HCC pathology. Any disagreements were resolved by consensus with a third senior pathologist.

If MVI was detected in any of the sampling protocols, the specimen was deemed positive for MVI. If multiple sections displayed the same MVI lesion (Figure 3A) or involved the same vessel (Figure 3B), it was recorded as a single MVI. The detection rate and quantity of MVI were recorded for each sampling protocol. The metastatic distance between each MVI lesion and the main tumor margin was measured and classified into five groups: ≤ 1.0 mm, 1.1-3.0 mm, 3.1-5.0 mm, 5.1-10.0 mm, and > 10.0 mm. The number of MVI detected by each sampling protocol in various metastatic distance groups was recorded.

Figure 3 Illustration of counting microvascular invasion-positive lesions. A: Microvascular invasion (indicated by blue arrows) in two sequential slices showing the same lesion, which was recorded as one lesion; B: Two microvascular invasion lesions (indicated by blue arrows) on one slice that invaded the same microvessel, which were recorded as one lesion.

Statistical analysis

The statistical analysis was conducted via SPSS 25.0 (IBM, Armonk, NY, United States). The normality of the quantitative data was assessed via the Shapiro-Wilk test. Descriptive statistics are presented as the means ± SD for continuous variables with a normal distribution and as the medians (ranges) for those with a skewed distribution. Paired t tests or paired Wilcoxon rank-sum tests were applied for paired data, whereas t tests or Mann-Whitney U tests were utilized for unpaired data based on the distribution of variables. Categorical variables are expressed as percentages, and the paired χ² test was used for paired data, whereas the χ² test was used for unpaired data with continuity correction based on theoretical frequencies. Statistical significance was defined as P < 0.05.

RESULTS

VX2-induced liver tumor formation in rabbits

In this study, the tumor inoculation process was well tolerated by all 40 rabbits without any associated mortality, implant-related infections, or behavioral changes. A total of 37 rabbits (92.5%) were implanted to establish the VX2 in situ liver tumor model. Implantation failed in three cases due to the transplantation of the tumor tissue into subcutaneous or mesentery tissue. Therefore, 37 rabbits were included in the analysis, and their baseline characteristics are shown in Table 1.

Table 1 Baseline characteristics of the rabbit liver tumor model, n (%).

Variables	Results
Rabbit liver tumor model
Success	37 (92.5)
Failure	3 (7.5)
Rabbit weight (kg), mean ± SD
Before tumor transplanting	2.5 ± 0.4
After successfully modeling	2.8 ± 0.4
Time of tumor growth (day), mean ± SD	20.4 ± 4.0
Tumor location
Left lateral lobe of liver	32 (80.0)
Left medial lobe of liver	5 (12.5)
Mesentery	2 (5.0)
Subcutaneous	1 (2.5)
Intrahepatic tumor number
Single	37 (100.0)
Multiple	0 (0.0)
Intrahepatic tumor size (cm), mean ± SD	1.9 ± 0.5
Microvascular invasion
Positive	30 (81.1)
Negative	7 (18.9)
Concomitant > 1 mm satellite nodule
Yes	9 (30.0)
No	21 (70.0)

The weights of the rabbits prior to tumor transplantation and after successful modeling were 2.5 ± 0.4 kg and 2.8 ± 0.4 kg, respectively, which were significantly different (P < 0.001). This finding suggested that the weight of the rabbits continued to increase normally after tumor inoculation. The growth of the implanted tumors was monitored for an average of 20.4 ± 4.0 days, during which all rabbits developed a single tumor with a maximum diameter of 1.9 ± 0.5 cm, primarily located in the left lateral lobe (32 cases) and left medial lobe (5 cases). All pathological sections were thoroughly and independently examined by two experienced hepatic histopathologists. During the examination, 56 MVI lesions had diagnostic disagreements, which were resolved through consensus with a third senior pathologist. Ultimately, 441 MVI lesions were determined in total, identifying 30 rabbits as positive for MVI, yielding an overall detection rate of 81% (30/37). Among these rabbits, nine had concurrent satellite nodules larger than 1 mm.

Baseline WSI characteristics

The study included 37 liver lobes to produce WSIs, each of which was sectioned into tissue blocks at 3 mm intervals, resulting in an average of 20.7 ± 2.3 blocks per lobe and a total of 765 blocks. Among these liver lobes, 22 lobes (458 blocks) were sectioned to generate a single WSI per block, resulting in 458 WSIs. The remaining 15 lobes (307 blocks) were sectioned at smaller intervals of 0.3 mm, yielding a median of 8 (range 7-9) WSIs per block and producing a total of 2441 WSIs. For MVI examination, protocol A included six regions per specimen, and a total of 222 regions were examined in 37 specimens, whereas protocols B, C, D, and E included 37, 202, 765, and 2441 WSIs, respectively.

Comparison of MVI detection rates with different sampling protocols

As the sampling range increased from protocols A to E, all MVI-positive cases in a previous protocol were included in the subsequent protocol. The MVI detection rates for protocols A, B, C, D, and E were 27% (10/37), 43% (16/37), 62% (23/37), 68% (25/37), and 93% (14/15), respectively. Figure 4A shows a sequential comparison of the MVI detection rates, which revealed that protocol B was superior to protocol A and that protocol C was superior to protocol B, with statistically significant differences observed (paired χ² test: P = 0.031, P = 0.016). Although protocol D had a greater detection rate than protocol C and protocol E had a greater detection rate than did protocol D, the differences were not significant (paired χ² test: P = 0.500; χ² test: P = 0.112). The results suggested that protocol C, which has a smaller sampling range, resulted in a greater MVI detection rate than the other protocols.

Figure 4 Impact of sampling protocol on the detection rate of microvascular invasion. A: Sequential comparisons of microvascular invasion detection rates were made for protocols A-E. Statistical analysis was conducted via paired χ² tests to compare protocol A vs B, B vs C, and C vs D, whereas the χ² test was used for comparing protocol D vs E; B: The optimal number of whole-mount slide images was further determined by comparing protocols B and C. Paired χ² tests were used for all comparisons, where whole-mount slide imaging WSI₁-WSI₄ correspond to the number of WSIs (1-4), respectively. MVI: Microvascular invasion; WSI: Whole-mount slide imaging.

We found that larger tumors produced a greater number of WSIs in protocol C, with a median of 5 WSIs (range 3-9) per tumor in this study. To determine the optimal number of WSIs needed for MVI detection, we incrementally increased the number of adjacent WSIs centered on the maximal tumor section in protocol B. Specifically, WSI₁-WSI₄ corresponded to WSIs 1-4, respectively. We found that the MVI detection rate was consistent with that of protocol C at WSI₄. As shown in Figure 4B, the MVI detection rate did not significantly differ between WSI₂ and WSI₁ (P = 0.250); however, the MVI detection rate of three consecutive interval WSIs at the tumor center (WSI₃) was significantly greater than that of WSI₁ (P = 0.031), whereas it did not significantly differ from that of WSI₄ (P = 1.000). Based on these findings, we recommend the use of WSI₃ for the optimal MVI detection rate, which includes 1 WSI of the maximal tumor section and 2 WSIs of adjacent sections.

Performance of different sampling protocols for MVI diagnosis

In this study, we assessed the performance of five protocols (A, B, C, D, and WSI₃) for detecting MVI in 37 specimens, with the overall MVI detection rate as the reference standard (Table 2). Protocol A (7-point sampling) exhibited an accuracy of only 46% and a high false-negative rate of 67%. Conversely, the accuracy of the WSI₃ protocol improved to 78%, and the false-negative rate decreased to 27%. Furthermore, protocol E had only 15 specimens, yet its detection rate reached 93% (14/15). The corresponding detection rates of protocols A, B, C, D, and WSI₃ were 40% (6/15), 53% (8/15), 60% (9/15), 60% (9/15), and 60% (9/15), respectively. In comparison, the detection rate of protocol E was significantly greater than that of each of the other protocols (paired χ² test: All P < 0.05). We evaluated the performance of other protocols for detecting MVI in 15 specimens using protocol E as the reference standard (Table 3). Similarly, protocol A was less accurate (47%) and had a greater false-negative rate (57%), whereas the WSI₃ protocol as more accurate (67%) and had a lower false-negative rate (36%). Our analysis indicated that WSI₃ emerged as the optimal protocol for detecting MVI because of its greater detection rate and narrower sampling range compared with the other protocols.

Table 2 Performance of protocols A, B, C, D, and three consecutive interval whole-mount slide images at the tumor center for detecting microvascular invasion in 37 specimens, using the overall microvascular invasion detection rate as the reference standard.

Metrics	Protocol A	Protocol B	WSI3	Protocol C	Protocol D
Positive rate	0.27 (10/37)	0.43 (16/37)	0.59 (22/37)	0.62 (23/37)	0.68 (25/37)
Accuracy	0.46 (0.42-0.75)	0.62 (0.45-0.77)	0.78 (0.61-0.90)	0.81 (0.64-0.91)	0.86 (0.70-0.95)
Sensitivity	0.33 (0.18-0.53)	0.53 (0.35-0.71)	0.73 (0.54-0.87)	0.77 (0.57-0.89)	0.83 (0.65-0.94)
Specificity	1.00 (0.56-1.00)	1.00 (0.56-1.00)	1.00 (0.56-1.00)	1.00 (0.56-1.00)	1.00 (0.56-1.00)
Positive predictive value	1.00 (0.66-1.00)	1.00 (0.76-1.00)	1.00 (0.82-1.00)	1.00 (0.82-1.00)	1.00 (0.83-1.00)
Negative predictive value	0.26 (0.12-0.47)	0.33 (0.15-0.57)	0.47 (0.22-0.73)	0.50 (0.24-0.76)	0.58 (0.29-0.84)
False negative rate	0.67 (0.47-0.82)	0.47 (0.29-0.65)	0.27 (0.13-0.46)	0.23 (0.11-0.43)	0.17 (0.06-0.35)

WSI₃: Three consecutive interval whole-mount slide images at the tumor center.

Table 3 Performance of protocols A, B, C, D, and three consecutive interval whole-mount slide images at the tumor center for detecting microvascular invasion in 15 specimens, using protocol E as the reference standard.

Metrics	Protocol A	Protocol B	WSI3/protocol C/protocol D
Positive rate	0.40 (6/15)	0.53 (8/15)	0.60 (9/15)
Accuracy	0.47 (0.22-0.73)	0.60 (0.33-0.83)	0.67 (0.39-0.87)
Sensitivity	0.43 (0.19-0.70)	0.57 (0.30-0.81)	0.64 (0.36-0.86)
Specificity	1.00 (0.05-1.00)	1.00 (0.05-1.00)	1.00 (0.05-1.00)
Positive predictive value	1.00 (0.52-1.00)	1.00 (0.60-1.00)	1.00 (0.63-1.00)
Negative predictive value	0.11 (0.01-0.49)	0.14 (0.01-0.58)	0.17 (0.01-0.64)
False negative rate	0.57 (0.30-0.81)	0.43 (0.19-0.70)	0.36 (0.14-0.64)

WSI₃: Three consecutive interval whole-mount slide images at the tumor center.

Comparison of MVI detection quantity and distribution with different sampling protocols

We evaluated the impact of different sampling protocols on the quantity and distribution of MVI lesions detected. The total numbers of detected MVI lesions via protocols A, B, C, D, and E were 15, 36, 107, 125, and 395, respectively, as illustrated in Figure 5A. These results show that expanding the sampling range significantly increased the quantity of MVI lesions detected (P < 0.05). Table 4 and Figure 5B compare the number of MVI lesions in different distance groups using various sampling protocols, indicating that most MVI lesions were found in adjacent nonneoplastic tissue, whereas fewer were detected in distant nonneoplastic tissue. All protocols showed a decreasing trend in the quantity of MVI lesions with increasing MVI metastatic distance. Moreover, protocol E had a significantly greater detection quantity of MVI than protocol D in each distance group (Mann-Whitney U test: P < 0.05). In addition, Table 4 and Figure 5C show the percentages of MVI lesions detected of in different distance groups for each sampling protocol. These results showed that, from protocols A to E, the percentage of MVI detected in adjacent (≤ 3 mm) nonneoplastic tissue gradually decreased from 93.3% to 51.7%, whereas the percentage in distant (> 3 mm) nonneoplastic tissue gradually increased from 6.7% to 48.3%. This finding suggests that expanding the sampling range can improve the ability to detect MVI in distant nonneoplastic tissue.

Figure 5 Quantity and distribution of microvascular invasion detected by different sampling protocols. A: Box plot showing the number of microvascular invasion (MVI) lesions detected by various sampling protocols. The Wilcoxon matched-pairs signed-rank sum test was employed to compare protocol A vs B, B vs C, and C vs D, whereas the Mann-Whitney U test was used to compare protocol D vs E; B: Trend of in the number of MVI lesion detected in different metastatic distance groups for each sampling protocol; C: Percentage distribution of the number of MVI lesions detected in different metastatic distance groups for each sampling protocol. MVI: Microvascular invasion; WSI: Whole-mount slide imaging.

Table 4 Comparisons of the number of microvascular invasion in metastatic distance groups using various sampling protocols, n (%).

MVI metastatic distance groups	Sampling protocol
	A	P value1	B	P value2	C	P value3	D	P value4	E
≤ 1.0 mm	12 (80.0)	0.004	22 (61.1)	< 0.001	64 (59.8)	1.000	64 (51.2)	0.005	118 (29.9)
1.1-3.0 mm	2 (13.3)	0.066	9 (25.0)	0.014	20 (18.7)	0.083	23 (18.4)	0.005	86 (21.8)
3.1-5.0 mm	0 (0.0)	0.083	3 (8.3)	0.041	12 (11.2)	0.102	16 (12.8)	0.006	45 (11.4)
5.1-10.0 mm	1 (6.7)	0.317	2 (5.6)	0.039	10 (9.4)	0.102	14 (11.2)	0.014	74 (18.7)
> 10.0 mm	0 (0.0)	1.000	0 (0.0)	0.317	1 (0.9)	0.066	8 (6.4)	0.008	72 (18.2)

¹The Wilcoxon matched-pairs signed-rank sum test was used to compare A vs B.

²The Wilcoxon matched-pairs signed-rank sum test was used to B vs C.

³The Wilcoxon matched-pairs signed-rank sum test was used to C vs D.

⁴The Mann-Whitney U test was used to compare D vs E.

MVI: Microvascular invasion.

DISCUSSION

Accurate and comprehensive histological diagnosis of MVI is crucial for expediting timely intervention and reducing recurrence in patients with HCC following surgery. Furthermore, the establishment of a preoperative MVI prediction model is highly contingent upon the precise histological identification of MVI after surgery. Therefore, improving the accuracy and reliability of postoperative pathological diagnosis is essential for preoperative evaluation and formulation of postoperative treatment strategies for MVI. In this study, an animal experiment was conducted to assess the reliability of the current pathological diagnosis of MVI. Furthermore, we used WSI-based sampling protocols that offer a more comprehensive MVI pathological diagnosis, especially in detecting distant peritumoral MVI lesions, thereby improving the detection rate. This approach improves the management of HCC patients with MVI and may reduce recurrence rates.

Consistent with previous investigations[15,20-22], our results revealed a relatively low detection rate (27%) of MVI using protocol A (7-point sampling). Protocol A exhibited an accuracy of only 46% and a notable false-negative rate of 67%, indicating the inadequacy of the current clinical sampling protocol in accurately detecting MVI in patients with HCC. Conversely, our study effectively demonstrated a significantly greater MVI detection rate when using protocol B (WSI sampling) than when using the 7-point protocol. This finding substantiates prior research that highlights the efficacy of WSI in increasing the detection rate of MVI compared with conventional pathological sampling methods[15].

Furthermore, we observed a noteworthy improvement in the MVI detection rate by increasing the number of WSIs obtained from tumor sections, as implemented in protocol C, compared to protocol B (62% vs 43%, P = 0.016). However, the detection rates of protocol C and protocol D did not significantly differ (62% vs 68%, P = 0.500), with protocol D involving the acquisition of a greater number of WSIs from distant nontumor sections. The lack of a significant difference may be because MVI mostly occurs around the tumor periphery, with a sparse distribution in distant nontumor tissue and a relatively small sample size. These findings suggest that increasing the number of WSIs in tumor sections provides a more effective approach for MVI detection than using nontumor sections. Notably, protocol E, which involved the acquisition of a substantial number of WSIs from 15 specimens, exhibited the highest MVI detection rate of 93% (14/15). Moreover, the detection rate of this protocol was significantly greater than that of each of the other protocols in these 15 specimens. The noteworthy detection rate of MVI achieved by protocol E emphasizes the potential risk of missed diagnoses associated with other sampling protocols. Additionally, we acknowledge the practical limitations of unrestricted sampling of HCC specimens in clinical practice. To optimize the diagnostic accuracy and alleviate the workload on pathologists, we identified WSI₃ as the optimal protocol, achieving an MVI detection rate comparable to that of protocol C with fewer samples. Overall, our study provides valuable insights for developing clinical sampling protocols based on WSIs to increase the accuracy of MVI diagnosis, which can notably impact clinical practice. In terms of diagnosis, it enables more precise tumor staging, which is crucial for formulating appropriate treatment plans. With respect to treatment decisions, more aggressive adjuvant therapies, such as targeted therapies or immunotherapies, may be considered. This strategy ultimately contributes to improved patient outcomes and reduced HCC recurrence rates.

In clinical practice, the number and distribution of MVI lesions play crucial roles in predicting postoperative recurrence and long-term survival in HCC patients[23,24]. Our study explored the quantity and distribution of MVI lesions and revealed a pattern in which the number of MVI lesions significantly increased with increasing sampling range and gradually decreased with increasing distance from the primary tumor, which was consistent with the occurrence and progression of tumor metastasis. Previous studies have reported that 93.4%, 93.9%, and 83.3% of peritumoral MVI lesions are distributed within 1 cm of the primary tumor according to the 7-point baseline protocol[14,25,26], but inadequate sampling via this protocol may lead to missed MVI detection in distant nonneoplastic tissue. Using the same protocol in the animal model, this study confirmed that 93.3% of MVI lesions were detected in liver tissue located < 3 mm from the tumor margin (similar to < 1 cm from the tumor in human HCC). However, the WSI pathology sampling protocol enabled an increased sampling range of distant nonneoplastic tissue and revealed an increasing percentage of MVI detection in > 3 mm distant nonneoplastic tissue from protocols B to E, with the highest percentage in protocol E (48.3%). This finding suggests that the previous 7-point sampling protocol may have overlooked MVI in distant nonneoplastic tissue, which could be a potential explanation for the high recurrence rate of HCC.

The present study utilized an animal model of rabbit VX2 in situ liver tumors, which may not be entirely generalizable to human subjects. However, we contend that this model is appropriate for several reasons. First, our study utilized WSIs of the entire liver lobe to examine the distribution of peritumoral MVI. This comprehensive approach allowed us to provide strong evidence to assess the accuracy of pathology sampling protocols and select the optimal pathological protocol to improve MVI detection. However, achieving such extensive evaluation in human liver tissue may be challenging; therefore, we selected the rabbit model as an alternative. Second, the rabbit VX2 in situ liver tumor model is widely used in liver cancer research and has blood supply features similar to those of human liver cancer[27]. Third, VX2 tumors in the rabbit liver have properties comparable to those of human liver cancer and present a high incidence of MVI. In a previous study by Wang et al[28], 14 rabbit VX2 in situ liver tumor animal models were constructed, and the incidence of MVI was 64% (9/14). In this study, the overall incidence of MVI reached 81% (30/37), which may be attributed to the broad range of pathological samples. Therefore, this model allows us to effectively replicate the occurrence and development of MVI in human liver cancer, which is essential for achieving the experimental objectives of our study. We acknowledge that a human histology study would have generated a stronger clinical message. However, given the difficulties associated with obtaining adequate human liver tissue samples and the limitations of using human tissue for comprehensive MVI distribution analysis, we contend that our use of a rabbit model was appropriate for our research.

Despite its notable findings, the present investigation is subject to several limitations. First, the small sample size may have contributed to the lack of significant differences between certain sampling protocols. Given the variability in MVI distribution, a larger sample size would likely increase the robustness and reliability of our findings regarding the performance of different sampling protocols in detecting MVI. Furthermore, the potential effect of tumor size on MVI detection rates across different protocols was not assessed by stratification. Finally, the optimal WSI₃ pathological protocol was not validated in human HCC. For human HCC, we recommend obtaining three consecutive WSIs at 1 cm intervals, which are sampled from the center of the tumor. In the future, we will validate the WSI₃ protocol in human HCC clinical settings to confirm its feasibility and practicality. Moreover, integrating artificial intelligence or machine learning techniques for automated WSI analysis will further optimize the operation process, facilitating the promotion of this valuable protocol in clinical practice.

CONCLUSION

In summary, our study underscores the inherent limitations of the current 7-point sampling protocol used in the pathological diagnosis of MVI, attributed to its low accuracy and high false-negative rate. However, we propose a practical and viable approach for pathological examination by obtaining three consecutive interval WSIs from the tumor center. This approach has demonstrated a remarkable increase in the detection rate of MVI and has successfully facilitated the ability to identify MVI in distant nonneoplastic tissue. Consequently, our findings provide valuable insights for optimizing pathological examination strategies, with the ultimate aim of enhancing MVI detection in future clinical practice.