In the current study, we examined a lot of patients to assess the clinical features of C-DFSP and FS-DFSP. Our results indicate no significant difference between patients with C-DFSP and FS-DFSP in terms of sex, age at presentation, age at the time of first diagnosis, interval from initial presentation to diagnostic confirmation, tumor size at the time of presentation, tumor size at the time of diagnosis, tumor growth, or annual tumor growth. Interestingly, compared with that of recurrent C-DFSP, the annual tumor growth of FS-DFSP was higher, although there was no significant difference.
Recent studies have revealed that the frequency of FS according to histopathology may be approximately 5% to 15% of all DFSP cases[12, 13]. Similar to previous studies, we found that FS-DFSP represented 11.7% of all DFSP cases. Connelly et al reported that the median age of FS-DFSP patients was significantly higher than that of C-DFSP patients. In the current study, we found that the median age of patients with FS-DFSP was only slightly higher than that of patients with C-DFSP. Many studies have reported that DFSP is generally diagnosed between the ages of 20 and 40 years. However, few studies have indicated the age at presentation of DFSP[6, 7, 22]. We found that there were some differences in the age at presentation of DFSP between the two groups. In our study, the peak incidence of FS-DFSP at presentation was observed in patients in their 20s, 30s, 40s and 50s, whereas the peak incidence of C-DFSP at presentation was observed in patients aged 12 years to the 6th decade of life.
DFSP can occur anywhere in the body. We found that FS-DFSP mostly occurred on the chest and posterior thighs, whereas C-DFSP mostly occurred on the chest. Currently, the correlation between DFSP incidence and sex remains unclear. Bowne et al. reported that the male-to-female ratio was nearly 1:1. Other studies reported a slight predominance of female patients.[4, 23] In the current study, however, we observed that the incidences of both FS-DFSP and C-DFSP were higher in males than in females. Correlations with prior trauma, surgical or burn scars, which had been reported in approximately 10% of DFSP cases, were unclear.[24, 25] In our series, trauma induced DFSP in 8.1% of patients.
Clinically, DFSP often presents as an indolent tumor. In the current study, we found that some lesions can be indolent, whereas they grow slowly or show rapid enlargement after a period of indolence. Interestingly, FS-DFSP had a significantly shorter time from indolence to rapid enlargement. FS changes have not been reported in children with DFSP[27-30]. Interestingly, the tumors of two patients with FS-DFSP presented in childhood. The age of one patient was 12 years, and that of the other was 15 years. There is evidence suggesting that FS-DFSP may be an evolution of C-DFSP. Previous studies have demonstrated that P53 and MDM2 are overexpressed in FS-DFSP. In addition, activation of Akt/mTOR, STAT3, ERK and PD-L1 may be related to the development or progression of DFSP[22, 31, 32].
Previously, the wide local excision (WLE) was the gold standard treatment for DFSP, with a recurrence rate ranging from 0% to 41%. Recently, Mohs micrographic surgery (MMS) has been proven to be an alternative to WLE that assesses 100% of the margins with maximum tissue conservation. Many studies comparing the recurrence rate of WLE and MMS for the treatment of DFSP have shown that the recurrence rate after MMS ranges from 0-6.7%[34-38]. Although the most adequate surgical method (i.e., MMS or WLE) for the treatment of DFSP remains controversial, there is evidence suggesting that MMS has lower rates of recurrence.[23, 34, 39] In some cases, DFSP might receive a simple excision because it is misdiagnosed as a benign mass, with high local recurrence (26-60%). FS-DFSP is highly aggressive and related to a high risk of local recurrence13,. FS changes can be commonly identified in primary tumors. In several studies, however, FS changes were detected only in recurrent tumors[12, 41]. Interestingly, our previous study showed that the proportion of FS-DFSP in the recurrent DFSP was higher than the primary DFSP. In a multicenter study, Eva A et al. revealed that after WLE, patients with FS-DFSP more often experienced recurrence than those with C-DFSP. In our recent study, we found that after MMS, FS change was an independent prognostic factor for local recurrence in both univariable and multivariable analyses.
It has been reported that 92-100% of DFSP cases usually show diffuse CD34 staining, can be positive for vimentin, nestin and apolipoprotein D, and can be negative for cytokerins, smooth muscle actin smooth muscle actin, S100, CD56, factor XIIIa, Stromelysin 3 and cathepsin K[24, 25, 42]. CD34 is reported to be negative in up to 50% of DFSP in FS-DFSP. In the current study, the negative ratio of CD34 in FS-DFSP was significantly lower than that in C-DFSP. Sasaki indicated that the Ki-67 index in FS-DFSP is significantly higher than the Ki-67 index in C-DFSP (C-DFSP: 8.9% vs FS-DFSP: 21.5%). Similarly, in our study, the average Ki-67 index in FS-DFSP cases was significantly higher than that in C-DFSP cases (C-DFSP: 8.1% vs FS-DFSP: 18.1%). As a nuclear protein, Ki-67 is related to ribosomal RNA synthesis and has an essential function in cell proliferation. Khor et al. indicated that a high index of Ki-67 in prostate cancer was related to an increased risk of distant metastasis, cancer-specific mortality and overall death. Several studies have shown that high Ki-67 levels were correlated with an obviously worse overall survival rate in mantle-cell lymphoma[46, 47]
The BP neural network is a kind of multilayer feedforward network that uses the error back-propagation algorithm. The BP neural network was first proposed by Pau1werboSS in 1974, but it has not been widely recognized. In the 1980s, Rumelhar et al renamed the BP algorithm, which was included in "Parallel Distributed Processing"[48-51]. Recently, the BP algorithm became the most widely used algorithm in neural networks. It was reported that approximately 90% of neural networks were based on the BP algorithm[19, 20]. At present, the BP neural network is widely used in disease recognition and diagnosis[19, 20]. In the present study, when the number of invisible layers is 10, the Levenberg-Marquardt algorithm can complete the learning of the entire training set sample size in 31 runs. The correct rates of classification and misdiagnosis were 84.1% and 15.9%, respectively. The classification accuracy and feasibility of the BP neural network model are high in FS-DFSP.
The retrospective nature of this research is the main limitation of the current study. In addition, long-term follow-up data were lacking in the current study. Nonetheless, this is one of the largest studies of DFSP, and despite its limitations, our study will provide valuable information to aid in clinical practice.