• computer modeling;
  • lymphatic anatomy;
  • lymphatic drainage;
  • Sappey’s lines;
  • skin;
  • symmetry


  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Author contributions
  8. References

Relatively little is known about the functional anatomy of the lymphatic vessels draining the skin. To address this issue, we previously created a three-dimensional computer model of skin lymphatic drainage, using melanoma lymphoscintigraphy (LS) data from 5232 patients. In this study we sought to extend our model by performing a detailed statistical analysis of the mapped LS data to characterize the functional anatomy of the superficial lymphatics without any a-priori spatial bias. We investigated the commonly held assumption that lymphatic drainage is symmetric between the two sides of the body. Results indicated that, with the exception of the lower anterior torso, posterior leg and a small section of the posterior torso, most skin regions with sufficient data showed symmetric drainage. LS data from each symmetric skin region were then reflected to the opposite side of the body to provide an increased LS dataset for subsequent analysis. Cluster analysis was then applied to this reflected LS dataset to group regions of skin that drained in a similar manner. Results defined nine large clusters of skin, largely draining to the dominant axillary, groin, cervical level II and preauricular node fields. Each of the four axillary and groin node fields defined large clusters of skin on the torso, dividing it into regions similar to the historical ‘Sappey’s lines’, although a fifth region of highly ambiguous drainage was also shown in the anterior and posterior center of the torso. Collectively, these results provide important new insights into skin lymphatic drainage, both improving and quantifying our understanding of functional lymphatic anatomy.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Author contributions
  8. References

A relatively small number of detailed studies have been conducted to characterize the human lymphatic system and very few have been carried out to investigate the lymphatic vessels draining the skin. One of the most comprehensive and influential studies was that by Sappey (1874), who investigated lymphatic drainage by injecting mercury into the interstitial tissues and lymphatic vessels of cadavers. He published his results in an extensive lymphatic atlas that contained a number of highly detailed anatomical drawings.

For over 100 years, the atlas of Sappey (1874) and his conclusions about lymphatic drainage were accepted as correct by the scientific and medical community. He claimed that lymphatic drainage from the skin of the trunk was symmetric between the two sides of the body, never crossing the vertical midline of the body or a theoretical horizontal line drawn around the waist. These lines were termed ‘Sappey’s lines’ and defined four zones of skin on the trunk, from which Sappey (1874) claimed that lymphatic drainage would occur to the corresponding axillary or groin node field.

The concepts of Sappey (1874) about skin lymphatic drainage went largely unchallenged until the 1970s when new information became available. Most of this new information came from lymphatic mapping studies in patients with melanoma using lymphoscintigraphy (LS) imaging, which is used to locate the lymph nodes draining a primary melanoma site on the skin (Uren et al. 1999). Sugarbaker & McBride (1976) showed that lymphatic drainage was unpredictable from a strip of skin 2.5 cm wide on either side of Sappey’s lines. These authors still maintained, however, that drainage from skin outside these zones would follow the original predictions of Sappey (1874) and occur to the axillary or groin node fields. Additional investigation over subsequent decades demonstrated further variability of skin lymphatic drainage that often contradicted the guidelines of Sugarbaker & McBride (1976) (Fee et al. 1978; Meyer et al. 1979; Sullivan et al. 1981; Bergqvist et al. 1984; Eberbach & Wahl, 1989). Norman et al. (1991) expanded the area of ambiguous drainage to include the head and neck and a much larger area of skin on the trunk, up to 11 cm on either side of Sappey’s lines.

More recently, LS studies conducted at the Sydney Melanoma Unit (SMU) have also shown that lymphatic drainage is highly variable between patients and that the guidelines of Sappey (1874) would predict drainage to the wrong node field in 30% of patients (Thompson & Uren, 2005). LS studies conducted at other melanoma treatment centers around the world have similarly shown that skin lymphatic drainage is clinically unpredictable (Leong et al. 2000; O’Toole et al. 2000; Statius Muller et al. 2002). This observed variability and clinical unpredictability highlight the need for a detailed statistical analysis of available data and pre-empt the need for a reclassification of the key anatomical features of functional lymphatic drainage.

Our previous work has involved mapping the SMU’s extensive LS database of over 5232 patients onto a 3D computer model of the skin and lymph nodes (Reynolds et al. 2007a). In brief, the skin model was constructed of 1098 finite elements (Bradley et al. 1997) using anatomical images from the Visible Human dataset (Spitzer et al. 1996). The SMU defined 43 separate node fields that directly drained the skin as listed in Table 1 and each of these node fields has also been modeled relative to the skin model using Visible Human images. Each of the primary melanoma sites in the LS database was then mapped onto the skin model using techniques previously reported by Reynolds et al. (2007a), giving the mapped model in Fig. 1, which shows the melanoma sites scaled according to their frequency. Corresponding draining node fields for each patient were mapped onto one or more of the model’s 43 node fields (Reynolds et al. 2007a).

Table 1.   Number of cases in the lymphoscintigraphy database that drained to each node field. Note that most node fields are located on both the left and right sides of the body.
Node fieldNo. of cases
Head and neck node fields
 Cervical level I86101
 Cervical level II274295
 Cervical level III8252
 Cervical level IV5846
 Cervical level V209201
 Supraclavicular fossa206193
Torso and upper limb node fields
 Triangular intermuscular space9597
 Internal mammary24
 Costal margin25
 Paravertebral or para-aortic36
 Upper mediastinal1
Lower limb node fields
Other node fields
 Interval node404

Figure 1.  Frequency of primary melanoma sites on the skin model.

Download figure to PowerPoint

Visualization of these mapped data showed that the most variable drainage patterns included skin on the head and neck (Reynolds et al. 2009), and skin on the torso close to Sappey’s lines (Reynolds et al. 2007b). This mapped model has provided the platform for the present study, where we have statistically analyzed patterns of skin lymphatic drainage. We investigated the last remaining assertions of Sappey (1874) that lymphatic drainage is symmetric between the two sides of the body. We then sought to functionally group regions of skin that drained in a similar manner, moving away from a-priori historical drainage assumptions towards a purely data-driven approach.

Materials and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Author contributions
  8. References

Symmetry of skin lymphatic drainage

To determine whether lymphatic drainage of the skin was symmetric between the two sides of the body, we compared LS data for primary melanomas located on either side of the coronal midline. Skin of the head and neck, torso, and upper and lower limbs was each divided into smaller regions for testing, to allow for a more refined analysis. Testing each region of skin provided one of three possible results: that lymphatic drainage was symmetric, that it was asymmetric or that there were insufficient data to draw a conclusion.

To test for symmetry, a generalized linear modeling (GLM) approach (Hardin & Hilbe, 2007) was used, implemented with the R statistical package (R Development Core Team, 2008). It was assumed that the LS data fit a multinomial distribution, which relates the independent variable, a patient’s primary melanoma site on the skin (i.e. the trial), to the dependent variable, one of 43 possible draining node fields (i.e. the outcome of that trial). Some patient cases in the LS database contained more than one draining node field and, as the multinomial distribution assumes that each trial has only one outcome, the LS data were modified so that cases with multiple node fields (shown in Fig. 2A) were separated into multiple data entries, one for each draining node field (shown in Fig. 2B).


Figure 2.  (A) Original lymphoscintigraphy (LS) data format: patient case 1 has drainage to one node field, case 2 has drainage to three node fields and case 3 has drainage to two node fields. (B) The LS data have been modified to give multiple data entries for patients with multiple node fields, so there are now six data entries.

Download figure to PowerPoint

A multinomial model was then fitted to this modified LS data, by using a link function and error distribution to relate the melanoma site locations (the predictor variables) to the probability of drainage to each node field (the dependent response variables). The melanoma site predictor variables were taken as the side of the body that each site was located on, the skin element that each site was located on and the interaction of these two terms. The unknown parameters of this model were then estimated using the maximum likelihood method, as described further by Faraway (2006).

After fitting the multinomial model, the Akaike information criterion (AIC) (Akaike, 1974) was used to determine the most important predictor variables:

  • image

where the loglikelihood gives a measure of the model fit and K is the number of model parameters. Predictor variables were chosen to minimize the AIC value, where the lowest AIC gave the best general model from amongst the candidates. If the side of the body was required as a predictor to give the lowest AIC value, then the LS data were defined to be asymmetric, as the drainage patterns from each side of the body were statistically different. Otherwise, if side was not an important predictor variable, then lymphatic drainage was considered to be symmetric.

The skin was divided into regions as follows. To maintain consistency with previous studies, the head and neck were divided into 12 skin regions according to Pathak et al. (2001), as shown in Fig. 3A. For the torso, the large amount of data available as well as the large number of possible draining node fields justified performing a highly spatially refined analysis of this region. Skin above the waist at the level of the umbilicus was divided into regions of 10 cm in height (shown in Fig. 3B), whereas the shoulder was treated as a separate region. The number of cases in each region on the posterior torso (excluding the shoulder) was greater than the number of cases on corresponding regions of the anterior torso. As there were sufficient data available, the upper regions of the posterior torso were divided further in the horizontal direction, resulting in approximately 10-cm square regions (shown in Fig. 3C). Meanwhile, the lower regions of the torso did not contain a sufficient amount of data to warrant further spatial discretization.


Figure 3.  (A) Head and neck skin regions, as defined by Pathak et al. (2001), and skin regions on the (B) anterior torso and (C) posterior torso.

Download figure to PowerPoint

The upper and lower limbs both contain two recognized lymph node fields. The upper limbs have axillary nodes in the armpits and epitrochlear nodes just above the elbow, whereas the lower limbs contain nodes in the groin and popliteal nodes behind the knee. Hence, the skin of each limb was divided into anterior and posterior skin regions, as well as upper and lower regions according to the location of the elbow and knee. Meanwhile, the hands and feet were each treated as separate skin regions.

During testing, the minimum sample size was arbitrarily set to 20 melanoma sites, to ensure that there were sufficient data in each skin region. LS data from patients who had had previous surgery were not included in this analysis, as these patients could have an altered lymphatic drainage pattern that would not be indicative of a normal population (Thompson et al. 2005). Also, drainage to interval nodes was not included during testing, as these nodes do not have a consistent anatomical location (Uren et al. 2000).

Cluster analysis

Cluster analysis was then performed to group skin elements with similar lymphatic drainage behavior without any a-priori bias. The clustering method that was used in this study was based on work by O’Sullivan et al. (1998), who implemented a clustering algorithm developed by Vössner & Braunstingl (1996).

To conduct cluster analysis on the LS data, the accumulated data contained on each skin element i were defined as the samples for classification (see Fig. 1), whereas the node fields j that drained each skin element were defined as characteristics of the sample. Each skin element had corresponding counts Xij to define the number of cases with drainage to each node field. These counts were then normalized to the range 0 ≤ Xij ≥ 1 by dividing the number of cases draining to each node field from each skin element by the total number of melanoma sites contained within that skin element. For example, a skin element containing six melanoma sites could have drainage from each site to the left groin node field, whereas two sites could also drain to the left popliteal node field. If characteristic X1 indicates drainage to the left groin and X2 indicates drainage to the left popliteal node field, and the remaining characteristics comprise the remaining node fields, then the normalized values for this skin element used for clustering were X11 = 1, X12 = 1/3 and X1j = 0 for j = 3..43.

Euclidean distances were then calculated between these normalized sample vectors to give a measure of the similarity of lymphatic drainage behavior, on which divisive hierarchical clustering was performed. For further details about this clustering algorithm see O’Sullivan et al. (1998).

Statistics were then determined to characterize the drainage behavior of each resultant cluster. The likelihood that lymphatic drainage will occur to a particular node field from each cluster was calculated and then non-parametric bootstrapping (Efron & Tibshirani, 1993) was implemented to find 95% confidence intervals for each of these likelihood values.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Author contributions
  8. References

Symmetry of skin lymphatic drainage

Figure 4 schematically shows the accumulated results of the symmetry analysis. Regions of skin that were considered to have symmetric lymphatic drainage are shown in skin color, symmetric regions are shaded in black, whereas regions with insufficient data are shaded red.


Figure 4.  Results of the symmetry analysis.

Download figure to PowerPoint

Head and neck

The LS data showed that lymphatic drainage for the head and neck was highly complex (Reynolds et al. 2009), occurring to a number of different node fields. Statistical analysis for each of the 12 head and neck regions (Fig. 3A) quantified that lymphatic drainage from most skin regions was symmetric. Details for each of these skin regions are given in Table 2, where the number of melanoma sites contained within each region from both the left and right sides of the body, and the number of individual data entries, are provided. The number of cases with sole drainage to interval nodes and individual data entries draining to interval nodes, which were not used during the analysis, are also given in brackets.

Table 2.   Results of the head and neck symmetry analysis. Cases with drainage to interval nodes that were not used in the analysis are given in brackets.
 Skin regionNo. of casesNo. of data entriesSignificant factorSymmetry identified
 Upper anterior3061 (5)Yes
 Lower anterior2870SideNo
 Upper lateral3577 (5)Yes
 Lower lateral4585 (4)Yes
 2Upper face
 Upper anterior55122 (8)Yes
 Lower anterior76 (1)159 (6)Yes
 Upper lateral57100 (6)Yes
 Lower lateral74 (1)144 (3)Yes
 3Lower face61128Yes
 4Upper anterior neck4999Yes
 5Lower anterior neck45 (1)Insufficient data
 6Coronal scalp2968 (4)Yes
 7Ear107190 (1)Yes
 8Upper coronal neck49101Yes
 9Lower coronal neck00Insufficient data
10Posterior scalp88202 (15)Yes
11Upper posterior neck71149 (2)Yes
12Lower posterior neck69136 (6)SideNo

Both the forehead and lower posterior neck showed asymmetry. The lower anterior neck and lower coronal neck did not contain sufficient data to be analyzed and have therefore not been assessed. The forehead showed that it had asymmetric drainage when treated as a whole region. As there were sufficient data from this region, it was divided further into four smaller regions to determine whether there was localized asymmetry. This division separated the skin into upper and lower anterior, and upper and lower lateral regions. Symmetry testing metrics indicated that three out of four of these forehead regions were symmetric, although analysis of the lower anterior region confirmed a difference between the drainage patterns of the two sides.

The upper face region was also divided into four separate regions, as again there were sufficient data to conduct a more spatially refined analysis. All of these regions had symmetric lymphatic drainage patterns. Meanwhile, asymmetry in the lower posterior neck corresponded with the highly complex drainage from this area of skin, where there were 16 possible draining node fields.


Skin of the torso contained a large proportion of the LS data, due in particular to a high frequency of primary melanoma sites located on the upper back (see Fig. 1). It was also one of the most variable regions for lymphatic drainage. Of the 43 different node fields draining the skin, 32 of them drained skin of the torso.

Statistical analysis results, as given in Table 3, indicated that lymphatic drainage from regions 1–4 (see Fig. 3B) on the anterior torso was symmetric, whereas region 5 and the lower anterior region were not symmetric. Anterior region 5 was located just above the umbilicus and had highly complex drainage with 10 different draining node fields. Asymmetry was also evident in the lower anterior torso region. This skin region covered a large area; however, as the data were sparse (visible in Fig. 1) it could not be divided further to allow a more refined analysis. There were more cases located proximally on the right side, which evidently caused the lymphatic drainage of this region to occur to the ipsilateral axilla more often than the left region (i.e. 32 vs. 6%). There were also more cases that drained to the contralateral groin from the right side than the left side, again probably due to the asymmetric location of the primary melanoma sites in this region.

Table 3.   Results of the torso symmetry analysis. Cases with drainage to interval nodes that were not used in the analysis are given in brackets.
Skin regionNo. of casesNo. of data entriesSignificant factorSymmetry identified
Anterior torso
 Region 160 (1)93 (2)Yes
 Region 282 (1)117 (8)Yes
 Region 333 (1)37 (3)Yes
 Region 478 (2)106 (14)Yes
 Region 575 (1)121 (16)SideNo
 Shoulder148187 (1)Yes
 Lower anterior3750 (1)SideNo
Posterior torso
 Region 1303651 (61)ElementYes
 Region 2351 (1)628 (41)Yes
 Region 3171 (1)272 (21)Yes
 Region 4130196 (25)Yes
 Region 598 (1)200 (24)ElementYes
 Region 6280 (1)393 (30)Yes
 Region 790106 (2)Yes
 Region 84446Yes
 Region 987 (2)100 (26)SideNo
 Region 1086 (1)146 (24)Yes
 Shoulder4045 (4)Yes
 Lower posterior5170 (7)Yes

Upon testing the posterior torso, nearly all regions showed symmetric lymphatic drainage behavior. The large number of cases on each side of the upper posterior torso draining to the ipsilateral axilla largely influenced this result. Region 9, however, indicated that it was asymmetric. This skin region drained to seven separate node fields, although four of these fields did not receive drainage from melanoma sites located on both sides of the body. Also, the low number of cases draining to the retroperitoneal and paravertebral node fields probably influenced this asymmetric result.

Upper and lower limbs

The entire upper limb showed that it had symmetric lymphatic drainage, owing primarily to its high likelihood of drainage to the ipsilateral axilla. In addition, both the hands and feet showed symmetric lymphatic drainage. In contrast, although the lower limbs always drained to the ipsilateral groin, they did not show complete symmetry. The posterior leg was not considered symmetric due to a different proportion of melanoma sites draining to the ipsilateral popliteal node fields. Twice as many of the melanoma sites located on the right leg drained to the right popliteal node field (25%) as opposed to melanoma sites on the left leg (12%). Table 4 details these results for each of the limbs.

Table 4.   Results of the upper and lower limb symmetry analysis. Cases with drainage to interval nodes that were not used in the analysis are given in brackets.
Skin regionNo. of casesNo. of data entriesSignificant factorSymmetry identified
Upper limb
 Anterior arm243257 (19)Yes
 Posterior arm161 (1)165 (13)Yes
 Anterior forearm288 (1)353 (32)Yes
 Posterior forearm5971 (6)Yes
Lower limb
 Anterior thigh317318 (3)Yes
 Posterior thigh102103 (3)Yes
 Anterior leg314325 (5)Yes
 Posterior leg307344 (6)SideNo
Hands4758 (4)Yes
Feet203238 (3)Yes

In summary, of the 5232 patients in the LS database, three patients with previous surgery and 17 patients with sole drainage to an interval node were excluded from the symmetry analysis. Of the remaining 5212 patients, 4605 were located on regions of skin considered to be symmetric, as detailed in Table 5. In order to provide a larger LS dataset for subsequent cluster analysis, melanoma sites within each of the symmetric skin regions were reflected to the opposite side of the body, giving a total of 9817 cases in a reflected LS database.

Table 5.   Number of cases contained in each skin region used for symmetry testing, along with the total number of cases in symmetric regions that were reflected. The number of cases with sole drainage to interval nodes are given in brackets.
Skin regionNo. of casesNo. reflected
Torso2244 (13)2045
Upper limbs751 (2)751
Lower limbs1040733
Head and neck927 (2)826
Total5121 (17)4605

Cluster analysis

Cluster analysis was then carried out using the reflected LS data. The advantage of this larger dataset was that it allowed for improved robustness in the definition of the boundaries of each resultant cluster. As for the symmetry testing, interval node data were not included during this analysis.

The hierarchical clustering algorithm was carried out on the reflected dataset, where initially all skin elements were contained in the same cluster. As the algorithm proceeded the skin elements were separated into smaller clusters. For some iterations, clusters or even individual skin elements dropped out of the analysis altogether.

The cluster algorithm iterations were stopped when they provided an anatomically reasonable result, which balanced the number of elements per cluster with the complexity of the results. At this particular iteration, nine large clusters and 10 small clusters had formed, where clusters containing more than five elements were considered large, whereas any clusters with less than five elements were considered small. It was observed that, beyond this iteration, the total number of clusters increased considerably, whereas the number of large clusters decreased. As shown on the skin model in Fig. 5, large clusters had formed on each lower limb, two large clusters were located on each side of the head, large clusters also formed on each side of the torso and upper limb, whereas another large cluster formed on both the anterior and posterior lower torso.


Figure 5.  Clustering results showing the functional division of the skin into regions with similar lymphatic drainage behavior.

Download figure to PowerPoint

Each of these final clusters corresponded with a particular pattern of lymphatic drainage. The statistics given in Table 6 indicate the draining node fields for each cluster, where only the dominant draining node fields are given. The dominant fields were the axilla, groin, cervical level II and preauricular node fields, which received drainage from 62, 29, 11 and 7% of patients in the LS database, respectively (Table 1).

Table 6.   Drainage statistics of the dominant draining node fields from each cluster.
ClusterLocationDraining node fieldsDrainage statistics (%) Mean (95% CI)
  1. R, right; L, left; C, cervical level; SC, supraclavicular fossa.

1Right lower limb (n = 1116)R groin100 (100, 100)
R popliteal6.36 (4.93, 7.89)
2Left lower limb (n = 1138)L groin99.74 (99.38, 100)
L popliteal4.92 (3.69, 6.24)
3Lower mid-torso (n = 111)R groin70.27 (62.16, 78.38)
L groin69.37 (60.36, 78.38)
R axilla18.02 (11.71, 25.23)
L axilla17.12 (10.81, 24.32)
4Right torso and upper limb (n = 2613)R axilla95.41 (94.56, 96.13)
L axilla16.80 (15.46, 18.29)
7Left torso and upper limb (n = 2636)L axilla95.33 (94.54, 96.09)
R axilla16.81 (15.36, 18.29)
9Right lateral neck and jawline (n = 100)R C274.00 (65.00, 82.00)
R C530.00 (21.00, 39.00)
R C327.00 (19.00, 36.00)
R C419.00 (12.00, 27.00)
R SC15.00 (8.98, 22.00)
10Right lateral scalp and cheek (n = 367)R C267.57 (62.67, 72.48)
R preauricular60.49 (55.31, 65.12)
R postauricular18.53 (14.44, 22.62)
R C112.81 (9.26, 16.36)
16Left lateral scalp and cheek (n = 375)L C267.47 (62.93, 72.27)
L preauricular62.13 (57.07, 67.47)
L postauricular17.87 (14.13, 21.60)
L C112.53 (9.33, 16.27)
17Left lateral neck and jawline (n = 78)L C270.51 (60.26, 79.52)
L C525.64 (16.67, 35.90)
L C325.64 (16.67, 35.90)
L C417.95 (8.97, 25.67)
L SC15.38 (7.69, 24.36)

Clusters 1 and 2 on the lower limbs indicated that 100 and 99.74% of all cases within these clusters drained to the right and left groin, respectively. A small number of melanoma sites located below the knee also drained to the popliteal node fields from these clusters. Clusters 4 and 7, located on each side of the torso and upper limbs, corresponded to a high likelihood of drainage to the ipsilateral axilla. Statistics showed that just over 95% of all melanoma sites located within these two clusters drained to the ipsilateral axilla, whereas nearly 17% of all sites drained to the contralateral axilla. Owing to the large amount of data in each of these clusters, the confidence intervals were small. Cluster 3, which was located in the middle of the lower anterior and posterior torso, was likely to drain to either groin node field (approximately 70%) and to either axillary node field (approximately 17%). This confirmed that there was a relatively large area of skin in the middle of the torso with ambiguous lymphatic drainage.

The remaining four clusters were located on either side of the head and grouped skin elements together on the lateral scalp and cheek (shaded orange and burgundy), and on the lateral neck and jaw line (shaded white and silver). Skin elements contained within the lateral scalp and cheek clusters always showed drainage to the preauricular and cervical level II node fields. According to the drainage statistics given in Table 6, ipsilateral cervical level II was the most common draining node field (approximately 68%), followed very closely by the preauricular node field (approximately 60%). Often drainage was also observed to the postauricular and other cervical level node fields.

Similar to the lateral scalp and cheek clusters, ipsilateral cervical level II was the most common node field draining the lateral neck and jaw line clusters. Over 70% of all melanoma sites from the clusters drained to this node field and all elements within these clusters also drained to cervical levels II, III and IV node fields. Often drainage also occurred to cervical level V and supraclavicular fossa nodes.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Author contributions
  8. References

The statistical methods that were implemented have allowed for a detailed quantitative analysis of skin lymphatic drainage. The mapped anatomical model has enabled knowledge of the superficial lymphatics to move beyond anatomically based definitions of drainage based largely on variants of Sappey’s lines towards a purely data-driven approach.

The implied and arguably final remaining assumption from the work of Sappey (1874), that skin lymphatic drainage is symmetric, has been tested. The results indicated that most skin regions did in fact show symmetric lymphatic drainage about the coronal midline of the body. Skin regions that displayed asymmetry included the lower posterior neck, lower anterior torso, posterior legs and small regions of the anterior forehead and posterior torso. It was apparent, however, that the asymmetry in these regions was probably due to an asymmetric distribution of melanoma sites, which would have influenced the draining node fields. The lower anterior neck and lower coronal neck could not be analyzed because there were insufficient data. Hence, it is possible that both the asymmetric regions and those without sufficient data would have been symmetric if more LS data had been available.

This is a significant result, as other studies have indicated asymmetry in the lymphatic system. Investigations into the distribution of lymph nodes on either side of the body have shown that there are more lymph nodes on the right side than the left side (Sapin, 1980). In addition, analyses on patients with breast cancer with axillary node drainage have shown that, although there are more axillary lymph nodes in the left side of the body, they are smaller in size than those on the right side (Capello et al. 2001; Dane et al. 2008). Further to this asymmetry of the lymphatic anatomy, it has also been shown that there is a left-sided lateralization in patients who develop cutaneous melanoma (Brewster et al. 2007).

A number of reasons have been put forward in order to explain these inherent asymmetries, including the embryological development of the lymphatic system and various genetic factors but a full explanation remains unclear. It is known that the lymphatic system begins to develop at 5 weeks gestation, in parallel to the blood vasculature system (Oliver, 2004). The lymphatic capillaries develop in a manner similar to blood vessels and major lymph channels usually follow the course of the main veins (Uren et al. 1999). Hence, it is likely that if the veins are symmetric then the lymphatic vessels will also be symmetric. Although the major veins in the limbs and head and neck are symmetric in topology, in the torso they are not entirely symmetric due to the asymmetric position of the heart. This could provide a tendency for the lymphatic vessels in the torso to follow an asymmetric pattern. Given this knowledge it is noteworthy that a large area of the torso showed symmetrical lymphatic drainage.

Although this study shows that the entire body may have symmetric lymphatic drainage, it is important to remember that these results are based on accumulated data. It is possible that individual patients may still have asymmetric drainage and nuclear medicine physicians and clinicians should continue to treat specific individuals with this in mind.

There were a number of aspects of the symmetry testing that could have been carried out differently, which may have altered the results. Primarily, division of the skin into separate regions was not a clearly defined procedure. The torso was first divided according to its anatomy, into anterior and posterior regions both above and below the umbilicus. Further division of the upper anterior torso was based on the geometrical boundaries defined on the skin model, which were arbitrarily defined during skin mesh construction (Reynolds et al. 2007a). Meanwhile, the upper and lower limbs were divided according to the lymphatic anatomy in these regions, even though there were enough data in some areas to support a more spatially refined analysis. A number of alternative skin divisions for each of these regions could have been used, which may have given different results. However, as a significant proportion of regions showed symmetry using different skin division methods, these results are likely to be robust.

It is also important to note that the modified LS data that were required to implement the GLM method meant that the data were no longer independent. It was possible to correct for this non-independence using generalized estimator equations (GEEs) (Hardin & Hilbe, 2002); however, the current library of GEE functions in the R statistical package was unable to handle the quantity of LS data in the database. Utilizing GEEs would have given the same model parameter estimates as a GLM, although the SEs for these estimates would have been inflated. This means that skin regions that were considered asymmetric using the GLM approach may have been considered symmetric using GEEs. Therefore, some of the asymmetric skin regions could in fact be symmetric if the data had been corrected for non-independence. The GLM approach was still considered appropriate, however, as it provided a conservative assessment of symmetry.

Subsequent cluster analysis has provided additional insight into regions of skin that showed functionally similar patterns of lymphatic drainage, based solely on the LS data. The results showed a clear anatomic division of the skin into nine separate clusters, which primarily grouped regions of skin according to the dominant draining node fields. Interestingly, the clusters draining primarily to axillary and groin node fields divided the trunk into regions comparable to Sappey’s lines. Even though there was variability of lymphatic drainage on the torso between individuals, Sappey’s lines appeared to conform to the most likely drainage behavior of these data. Cluster 3, however, which formed in the center of the anterior and posterior torso (shown in Fig. 5), clearly demonstrated that there was still a significant region of skin with ambiguous drainage to axillary and groin node fields. Note that the sparsity of data available in the anterior groin region (shown in Fig. 1) accounts for considerable asymmetry in the upper boundaries of the groin clusters. These results can be directly compared with the heat maps that we presented in our previous work (Reynolds et al. 2007b), which visualized the likelihood that the skin drained to the axillary or groin node fields. Regions of skin displayed on the heat maps that showed approximately 100% likelihood of drainage to the axillas or groin node fields largely comprised the axillary and groin clusters.

As with symmetry testing, the cluster analysis also had limitations. The cluster algorithm grouped LS data according to elements on the skin model. The advantage of this approach was that it was computationally straightforward to implement; however, it also meant that the data were homogenized across elements that were arbitrarily chosen during skin mesh construction. Restriction of the boundaries of each cluster to the boundaries of the skin elements was another limitation. In an ideal situation, uniformly sized skin regions would have been used rather than skin elements that have large variations in size. In addition, there would be adequate LS data to provide probabilities that represent the entire population. A number of elements on the skin mesh did not have any data present and therefore could not be grouped in a cluster. To enable a comprehensive cluster analysis, more LS data would be required.

Although each of the statistical tests that were used had inherent limitations, they have provided important new insights into skin lymphatic drainage. Significantly, it has been demonstrated that lymphatic drainage of the skin is likely to be entirely symmetric. The cluster analysis has clearly defined areas of skin that nearly always showed drainage to the ipsilateral axilla, groin, cervical level II and preauricular node fields. In addition, the drainage statistics and associated confidence intervals that were calculated have provided quantitative information about the functional anatomy of the superficial lymphatics that was previously unknown.

Author contributions

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Author contributions
  8. References

R.F.U. and J.F.T. provided access to the SMU’s LS database, which had been collected and recorded by R.F.U. N.P.S. and P.R.D. obtained funding and gave study supervision. C.G.W. conceived the approach and methods for the symmetry testing, whereas M.J.O.’S. provided the methods for the cluster analysis. H.M.R. created the anatomical model, carried out the statistical analysis and wrote the initial draft of the report. All authors provided critical revision of the manuscript and approval of the article.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Author contributions
  8. References
  • Akaike H (1974) A new look at the statistical model identification. IEEE Trans Automa Contr 19, 716723.
  • Bergqvist L, Strand SE, Hafström L, et al. (1984) Lymphoscintigraphy in patients with malignant melanoma: a quantitative and qualitative evaluation of its usefulness. Eur J Nucl Med 9, 129135.
  • Bradley CP, Pullan AJ, Hunter PJ. (1997) Geometric modelling of the human torso using cubic hermite elements. Ann Biomed Eng 25, 96111.
  • Brewster DH, Horner MJ, Rowan S, et al. (2007) Left-sided excess of invasive cutaneous melanoma in six countries. Eur J Cancer 43, 26342637.
  • Capello F, Bellafiore M, Palma A (2001) Study of axillary lymph node asymmetry in a female population. J Anat 199, 617620.
  • Dane S, Yildirim S, Koc M, et al. (2008) Asymmetries in breast cancer lateralization and both axillary lymph node number and metastatic involvement. Lymphology 41, 7579.
  • Eberbach MA, Wahl RL (1989) Lymphatic anatomy: functional nodal basins. Ann Plast Surg 22, 2531.
  • Efron B, Tibshirani RJ (1993) An Introduction to the Bootstrap. New York: Chapman & Hall.
  • Faraway JJ (2006) Extending the Linear Model with R: Generalized Linear, Mixed Effects and Nonparametric Regression Models. Florida: Chapman and Hall/CRC.
  • Fee HJ, Robinson DS, Sample WF, et al. (1978) The determination of lymph shed by colloidal gold scanning in patients with malignant melanoma: a preliminary study. Surgery 84, 626632.
  • Hardin JW, Hilbe JM (2002) Generalized Estimating Equations. Florida: Chapman & Hall/CRC.
  • Hardin J, Hilbe J (2007) Generalized Linear Models and Extensions. College Station, Texas: Stata Press.
  • Leong SP, Achtem TA, Habib FA, et al. (2000) Discordancy between clinical predictions vs lymphoscintigraphic and intraoperative mapping of sentinel lymph node drainage of primary melanoma. Arch Dermatol 136, 12691271.
  • Meyer CM, Lecklitner ML, Logic JR, et al. (1979) Technetium-99m sulfur-colloid cutaneous lymphoscintigraphy in the management of truncal melanoma. Radiology 131, 205209.
  • Norman J, Cruse W, Espinosa C, et al. (1991) Redefinition of cutaneous lymphatic drainage with the use of lymphoscin-tigraphy for malignant melanoma. Am J Surg 162, 432437.
  • O’Sullivan M, Vössner S, Wegener J (1998) Testing Temporal Correctness of Real-Time Systems - A New Approach Using Genetic Algorithms and Cluster Analysis. 6th European International Conference on Software Testing, Analysis and Review (EuroSTAR ‘98). Munich, Germany: EuroSTAR.
  • O’Toole GA, Hettiaratchy S, Allan R, et al. (2000) Aberrant sentinel nodes in malignant melanoma. Br J Plast Surg 53, 415417.
  • Oliver G (2004) Lymphatic vasculature development. Nat Rev Immunol 4, 3545.
  • Pathak I, O’Brien CJ, Petersen-Schaeffer K, et al. (2001) Do nodal metastases from cutaneous melanoma of the head and neck follow a clinically predictable pattern? Head Neck 23, 785790.
  • R Development Core Team (2008) R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing.
  • Reynolds HM, Dunbar PR, Uren RF, et al. (2007a) Mapping melanoma lymphoscintigraphy data onto a 3D anatomically based model. Ann Biomed Eng 35, 14441457.
  • Reynolds HM, Dunbar PR, Uren RF, et al. (2007b) Three-dimensional visualisation of lymphatic drainage patterns in patients with cutaneous melanoma. Lancet Oncol 8, 806812.
  • Reynolds HM, Smith NP, Uren RF, et al. (2009) Three-dimensional visualisation of skin lymphatic drainage patterns of the head and neck. Head Neck 31, 13161325.
  • Sapin MR (1980) Human lymph node asymmetry and its applied importance. Arkhiv Anat Gistol Embriol 79, 5863.
  • Sappey MPC (1874) Anatomie, Physiologie, Pathologie des Vaisseaux Lymphatiques consideres chez L’homme at les Vertebres. Paris: A. Delahaye et E. Lacrosnier.
  • Spitzer VM, Ackerman MJ, Scherzinger AL, et al. (1996) The visible human male: a technical report. J Am Med Inform Assoc 32, 118130.
  • Statius Muller MG, Hennipman FA, Van Leeuwen PA, et al. (2002) Unpredictability of lymphatic drainage patterns in melanoma patients. Eur J Nucl Med Mol Imaging 29, 255261.
  • Sugarbaker EV, McBride CM (1976) Melanoma of the trunk: the results of surgical excision and anatomic guidelines for predicting nodal metastasis. Surgery 80, 2230.
  • Sullivan DC, Croker BP, Harris CC, et al. (1981) Lymphoscintigraphy in malignant melanoma: 99mtc antimony sulfur colloid. Am J Roentgenol 137, 847851.
  • Thompson JF, Uren RF (2005) Lymphatic mapping in management of patients with primary cutaneous melanoma. Lancet Oncol 6, 877885.
  • Thompson JF, Uren RF, Saw RP, et al. (2005) Internal mammary node metastasis from primary cutaneous melanoma on the anterior chest. ANZ J Surg 75, 723725.
  • Uren RF, Thompson JF, Howman-Giles R (1999) Lymphatic Drainage of the Skin and Breast: Locating the Sentinel Nodes. Amsterdam: Harwood Academic.
  • Uren RF, Howman-Giles R, Thompson JF, et al. (2000) Interval nodes: the forgotten sentinel nodes in patients with melanoma. Arch Surg 135, 11681172.
  • Vössner S, Braunstingl R (1996) G.O.A.L. (Genetic Optimization Algorithm). Austria: Genetic Optimization Lab, Technical University Graz.