The treatment of choice for PMME was surgical resection with dissection of the lymph nodes [19]. Because of the propensity for submucosal spread of PMME, the surgical procedure should include a radical procedure with a greater margin than that used for the usual SCC [16], and in PMME, aggressive lymph node dissection was beneficial for accurate staging, potentially reducing recurrence and improving survival [20, 21]. So, it was important to obtain a preoperative diagnosis of PMME for choosing more aggressive surgery. This study found that thickness of esophageal tumor, AP CT value mean, and tumor range showed significant difference between PMME and SCC. This diagnosing model yielded perfect performance for distinguishing PMME and SCC. To the best of our knowledge, few studies focused on CT analysis and pathologic comparison for diagnosing PMME. To date, the present study was the largest population of PMME patients regarding radiological finding with more than 10 years span.

In this study, we found that PMME tended to appear intraluminal expansively growth pattern, which meant a broad base polypoid appearance mass protruding to the lumen with well-circumscribed boundary and smooth surface. In contrast, diffuse thickening and infiltrative growth pattern were more likely to occur in SCC lesions. It was consistent with other reports [16, 17]. Previous report found that in PMME, ulcerations may be present, but more often the tumor was covered with intact mucosa [16]. We also found that there was no difference in the presence of ulceration between PMME and SCC. The superficial ulcer was commonly observed in PMME, while deep ulcer due to tumor infiltrating to the esophageal wall was often presented in SCC. Small focal necrosis was a common manifestation in both PMME and SCC, which could not be used as the distinguishing point for this differentiation.

Our study showed that the thickness of PMME was higher than that of SCC and was one of the independent parameters for diagnosing PMME. The expansive growth type may result in the high size in PMME. The origination from basal layer, and expansive growth type in PMME may contribute to less esophageal symptoms and result in the higher size at initial diagnosis. Significant differences in CT value-related parameters were AP CT value mean and max between PMME and SCC in univariable analysis. We also found that PMME more often presented obvious enhancement in arterial phase, and SCC was prone to show progressive enhancement and enhanced peak in delayed phase. Tang et al. also found that enhanced CT scanning revealed obvious lesion enhancement, suggesting a rich tumor blood supply [18]. The other previous reports also found that contrast-enhanced CT showed malignant melanoma uneven enhanced in liver and stomach [22, 23]. In our study, interstitial vessels in PMME were more abundant than that of SCC in CD34 immunohistochemical staining. So, more interstitial vessels in PMME contributed to obvious enhancement in arterial phase. Rich fibrosis component in SCC was associated with peak enhancement in delayed phase.

Leiomyoma was the most common benign tumors of the esophagus [24]. The majority of esophageal leiomyoma originated from the inner circular muscle and two-thirds of this tumor was located in the middle or low thoracic esophagus [8]. Sometimes, PMME may be misdiagnosed as esophageal leiomyoma. In this study, we found that compared with PMME, esophageal leiomyoma often showed a smoothly marginated homogeneous mass with slightly homogeneous enhancement in arterial and delayed phases. Metastatic lymph nodes were not observed in esophageal leiomyoma. These results in our study were consistent with that of previous studies [8, 25]. In addition, ΔAP-N CT value max of tumor was an independent factor for differentiating PMME from esophageal leiomyoma. The AUC of ΔAP-N CT value max larger than the cutoff value of 16HU for diagnosing PMME from leiomyoma was 0.929. The robustness of our study was the use of quantitative parameter for this differentiation with high efficiency.

Even if the endoscopic biopsy was used, the preoperative diagnostic accuracy of PMME was only about 80% [4]. Sometimes the histological diagnosis of PMME can be challenging for several reasons. First, approximately 10–25% of PMME cases presented various colors, such as purple, brown, and white, depending on the melanin quality [5]. Second, when biopsy specimen lacked of melanin granules, the tumor might be incorrectly diagnosed as an epithelial carcinoma. Third, because of PMME arising from the melanocytes in the basal layer of squamous epithelium, the superficial biopsy may produce negative result. This study showed that morphological characteristics combined with CT value parameters were effective in diagnosing PMME.

In terms of clinical use, the radiologists could consider the result of diagnostic models in diagnosing PMME according to the probability of PMME from 0 to 1. The parameters of an esophageal tumor were applied to the two models, respectively. If there was a difference in diagnosis of PMME between the two models, the subjective diagnosis was needed to correct the diagnosis by combining radiological findings and clinical features. The differentiation between SCC and esophageal leiomyoma was evaluated by radiologists, considering the clinical and radiological features. When esophageal tumors were difficult to diagnose by radiologists assisted by our models, we recommend further evaluation with endoscopic ultrasound (EUS) and fine-needle aspiration (FNA) to look for the presence of PMME. When our CT analysis suggested a diagnosis of PMME for the esophageal lesion, but the histological diagnosis cannot identified in the initial biopsy, the immunohistochemical analysis of biopsy specimen or a deeper esophageal biopsy should be performed to obtain an accurate diagnosis.

There were several limitations in this study. First, it was a single-center retrospective study, and sample ratio might be biased. Second, because of the rarity, our study only included a small number of patients of PMME; a much larger database from multicenter with considerably larger sample was needed to validate the robustness and reproducibility of our result. Third, not all patients received surgery and the resected specimens were not allowed for pathological analysis because the PMME had aggressive biological behavior and these patients without receiving surgery were diagnosed late. Fourth, among these non-epithelial neoplasms, esophageal leiomyoma was the most common mesenchymal tumor of esophagus, unlike in the gastrointestinal tract, where GISTs predominated; the other non-epithelial neoplasms were extremely rare [8]. So, we only chose esophageal leiomyoma as a control group to differentiate from PMME. Searching for the radiologic finding of the other non-epithelial neoplasms is another study.

Comparing with SCC, PMME was prone to manifest expansive growth pattern, more arterial enhancement, larger tumor thickness, and less infiltrative range. This CT model might hold promise in discriminating PMME from SCC. Comparing with leiomyoma, PMME showed more obvious heterogeneous enhancement pattern, the presence of malignant lymph nodes, necrosis in the tumor, and non-smooth surface of tumor. These CT models might hold promise in discriminating PMME from SCC and esophageal leiomyoma. Although PMME was less common than SCC, radiologists should be familiar with the imaging and pathologic features of PMME as well as their malignant behavior and appropriate patient management.