Dai-Wei Liu

Background: Head and neck cancer is highly prevalent in Taiwan. Its treatment mainly relies on clinical staging, usually diagnosed from images. A major part of the diagnosis is whether lymph nodes are involved in the tumor. We present an algorithm for analyzing clinical images that integrates a deep learning model with image processing and attempt to analyze the features it uses to classify lymph nodes. Methods: We retrospectively collected pretreatment computed tomography images and surgery pathological reports for 271 patients diagnosed with, and subsequently treated for, naïve oral cavity, oropharynx, hypopharynx, and larynx cancer between 2008 and 2018. We chose a 3D UNet model trained for semantic segmentation, which was evaluated for inference in a test dataset of 29 patients. Results: We annotated 2527 lymph nodes. The detection rate of all lymph nodes was 80%, and Dice score was 0.71. The model has a better detection rate at larger lymph nodes. For those identified lymph nodes, we found a trend where the shorter the short axis, the more negative the lymph nodes. This is consistent with clinical observations. Conclusions: The model showed a convincible lymph node detection on clinical images. We will evaluate and further improve the model in collaboration with clinical physicians.

There have been major developments in deep learning in computer vision since the 2010s. Deep learning has contributed to a wealth of data in medical image processing, and semantic segmentation is a salient technique in this field. This study retrospectively reviews recent studies on the application of deep learning for segmentation tasks in medical imaging and proposes potential directions for future development, including model development, data augmentation processing, and dataset creation. The strengths and deficiencies of studies on models and data augmentation, as well as their application to medical image segmentation, were analyzed. Fully convolutional network developments have led to the creation of the U-Net and its derivatives. Another noteworthy image segmentation model is DeepLab. Regarding data augmentation, due to the low data volume of medical images, most studies focus on means to increase the wealth of medical image data. Generative adversarial networks (GAN) increase data volume via deep learning. Despite the increasing types of medical image datasets, there is still a deficiency of datasets on specific problems, which should be improved moving forward. Considering the wealth of ongoing research on the application of deep learning processing to medical image segmentation, the data volume and practical clinical application problems must be addressed to ensure that the results are properly applied.

Purpose Breast-conserving surgery (BCS) followed by whole breast radiation therapy (BCS-WBRT) or total mastectomy without WBRT (TM-no-WBRT) is the primary treatment for early stage breast cancer patients. Our study aimed to identify which early stage breast cancer treatment strategies had a subsequent lower incidence rate of mood disorder over a period of 10 years after the primary treatment. Methods This retrospective cohort study consisted of newly diagnosed early stage breast cancer patients in Taiwan from 2000 to 2013 using the National Health Insurance Research Database in Taiwan. We used a 1:1 propensity score matching by age to enrol patients into the BCS-WBRT and TM-no-WBRT groups. Statistical analyses were performed to calculate the hazard ratio and cumulative incidence rate. Results Our study consisted of 876 BCS-WBRT patients and 1949 TM-no-WBRT patients. After propensity score matching, each study group included 876 patients. The results showed that the mood disorder incidence rate was lower in the BCS-WBRT group than in the TM-no-WBRT group. Multivariate Cox regression analysis revealed that the BCS-WBRT group had a decreased risk of developing mood disorder (adjusted hazard ratio 0.69, 95% CI 0.53–0.90, p < 0.01). Furthermore, the Kaplan–Meier analysis showed that the BCS-WBRT group had a lower cumulative incidence rate of mood disorder, especially depression, after undergoing 10 years of primary treatment (p = 0.004). Conclusion Our results indicated that BCS-WBRT was associated with a lower risk of development of mood disorder over a 10-year period compared to TM-no-WBRT in early stage breast cancer patients. Our findings may provide helpful information, along with other clinical data, for breast cancer patients as they choose the type of appropriate surgery for treatment.

Z-ligustilide (or ligustilide) is found in Angelica sinensis (Oliv.) Diels and may exert potential benefits in cancer treatment. Previous research has reported that ligustilide has anti-cancer effects on several types of cancer cells. However, studies of ligustilide on oral cancer cells have not been reported, especially under hypoxic conditions. This study focuses on the molecular mechanism of ligustilide-induced apoptosis in hypoxic oral cancer cells. We found that in hypoxic TW2.6 cells, ligustilide inhibited cell migration and induced caspase-dependent apoptosis. Accumulation of c-Myc accompanied by BH3-only members suggests that ligustilide may induce c-Myc-dependent apoptosis. In addition, we reported that ligustilide has an effect on ER-stress signaling. By using inhibitors of c-Myc, IRE1α, and ER-stress inhibitors, we found that cell morphologies or cell viability were rescued to some degree. Moreover, ligustilide is able to increase the expression of γ-H2AX and enhance the occurrence of DNA damage in oral cancer cells after radiation treatment. This result suggests that ligustilide has potential as a radiation sensitizer. Altogether, we propose that ligustilide may induce c-Myc-dependent apoptosis via ER-stress signaling in hypoxic oral cancer cells.

We investigated whether the combination of primary tumor and nodal ¹⁸F-FDG PET parameters predict survival outcomes in patients with nodal metastatic non-small cell lung cancer (NSCLC) without distant metastasis. We retrospectively extracted pre-treatment ¹⁸F-FDG PET parameters from 89 nodal-positive NSCLC patients (stage IIB–IIIC). The Cox proportional hazard model was used to identify independent prognosticators of overall survival (OS) and progression-free survival (PFS). We devised survival stratification models based on the independent prognosticators and compared the model to the American Joint Committee on Cancer (AJCC) staging system using Harrell’s concordance index (c-index). Our results demonstrated that total TLG (the combination of primary tumor and nodal total lesion glycolysis) and age were independent risk factors for unfavorable OS (p < 0.001 and p = 0.001) and PFS (both p < 0.001), while the Eastern Cooperative Oncology Group scale independently predicted poor OS (p = 0.022). Our models based on the independent prognosticators outperformed the AJCC staging system (c-index = 0.732 versus 0.544 for OS and c-index = 0.672 versus 0.521 for PFS, both p < 0.001). Our results indicate that incorporating total TLG with clinical factors may refine risk stratification in nodal metastatic NSCLC patients and may facilitate tailored therapeutic strategies in this patient group.

Sebaceous carcinoma is a rare malignancy, accounting for 1–5.5% of eyelid malignancy. When local excision is even not feasible, definitive radiotherapy (RT) may be considered as an alternative. An 84-year-old woman, with initial presentation of right eye pain, suffered from a papillary tumor on right upper eyelid, 3.5 cm × 0.6 cm in size. Excision biopsy revealed sebaceous carcinoma. Image studies excluded regional and distant metastasis. Definitive RT with 54 Gy, 1.8 Gy per fraction, 5 days per week, using tungsten eye shield was delivered. The radiation dose of cornea, lens and retina was compared to the dose constraint recommended by studies from Radiation Therapy Oncology Group (RTOG). We used 6 cm × 6 cm cone, with 6 MeV electron beam irradiation, water phantom and ion chamber to measure the dose. The daily dose after using eye shield was 14.4, 35.9 and 1.7 cGy to cornea, lens and retina, respectively, compared to 210.8, 231.5 and 8.03 cGy without eye shield use, respectively. Eye shield masked the radiation dose about 93.2% and 79.9% to cornea and lens, respectively. Eye shield could mask extra radiation to cornea and retina and should be used for RT of eyelid tumor.

The 5-year overall survival rates for head and neck cancer (HNC) relies on distant metastasis. Importantly, the epithelial-mesenchymal transition (EMT) is believed to be an initial step of metastasis. However, the relationship of epigenetic with EMT formation is still unexplored in HNC. This study focuses on invasive subclones of HNC cell lines through the simulation of invasion in vitro; and underlying mechanisms were analyzed including DNA methylation and gene expression profile. Invasive subclones of NHC cell lines were successfully obtained using transwell coated with Matrixgel. Cells invaded through 8 μm pore several times were subcultured and examined with EMT features including morphology, EMT marker genes expression, and invasive ability. Moreover, compared the profile of genes expression in parental and invasive cells was analyzed using mRNA expression array. DNA methyltransferase 3B (DNMT 3B) was upregulated in invasive subclones and might control the 5′ region of E-cadherin (E-cad) methylation and further inhibited E-cad protein expression. Interference of DNMT 3B by siRNA or miRNA 29b could reduce EMT and cell invasion. Expression array analysis revealed the most possible involved pathways in cell invasion including arginine and proline metabolism, TGF-beta, and focal adhesion. DNMT 3B might control EMT by DNA methylation manner in invasive HNC cell lines. Moreover, miR-29b mimic downregulated DNMT 3B and inhibited EMT and cell invasion indicated the role of therapeutic agent for invasive HNC. Genes identified from array data and new molecules are involved in metastasis of HNC need further validation.

Aberrant insulin-like growth factor-binding protein 7 (IGFBP-7) expression has been found in various cancers such as prostate, breast, and colon. IGFBP-7 induced the apoptosis of tumor and potentially predicted the clinical outcome in some cancers is further demonstrated. This study investigates the causes and underlying mechanisms of aberrant IGFBP-7 expression in unravelling head and neck squamous cell carcinoma (HNSCC). A total of 47 oral tongue cancer patient samples were primarily analyzed for the methylation status in 5′ region of IGFBP-7 by methylation-specific PCR (MS-PCR). Subsequently the invasion, overexpression, and knockdown of IGFBP-7 in the HNSCC A253 invasive subpopulation were employed to examine the effect of IGFBP-7. The epithelial–mesenchymal transition (EMT) marker genes and AKT/GSK3β/β-catenin signaling were further evaluated by Western blot for the understanding the role of aberrant IGFBP-7 expression and thereof putative mechanism. EMT expressed in the invasive subpopulation of HNSCC cell lines (A253 and RPMI 2650) was contemporary with the down-regulation of IGFBP-7. After treatment with 5-AZA-2′ deoxycytidine, the de-methylated CpG sites in the 5′ region of IGFBP-7 were observed and IGFBP-7 mRNA expression was also restored. Accordingly, re-expression IGFBP-7 in invasive subpopulation of A253 could induce the mesenchymal–epithelial transition (MET) and concurrently inhibited the cell invasion. Moreover, IGFBP-7 methylation status of 47 oral tongue tumors showed a positive correlation to invasive depth of the tumor, loco-regional recurrence, and cancer sequence. IGFBP-7 can alter EMT relative marker genes and suppress cell invasion in A253 cell through AKT/GSK3β/β-catenin signaling. The epigenetic control of IGFBP-7 in the invasion and metastasis of HNSCC was reported, suggesting that IGFBP-7 could be a prognostic factor for the probability of invasion and a therapeutic remedy.

The 2010 Revisions to the McDonald Criteria have established that dissemination in time (DIT) of multiple sclerosis (MS) can be demonstrated by simultaneous presence of asymptomatic gadolinium-enhancing and nonenhancing lesions on a single magnetic resonance imaging (MRI). However, gadolinium-based contrast agents (GBCAs) have contraindications. Diffusion-weighted imaging (DWI) can detect diffusion alterations in active inflammatory lesions. The purpose of this study was to investigate if DWI can be an alternative to contrast-enhanced T1-weighted imaging (CE T1WI) for demonstrating DIT in MS. We selected patients with clinically definite MS and evaluated their baseline brain MRI. Asymptomatic lesions were identified as either hyperintense or nonhyperintense on DWI and enhancing or nonenhancing on CE T1WI. Fisher’s exact test was performed to determine whether the hyperintensity on DWI was related to the enhancement on CE T1WI (P < 0.05). The sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy of the DWI to predict lesion enhancement were calculated. Twenty-two patients with 384 demyelinating lesions that were hyperintense on T2-weighted imaging and more than 3 mm in size were recruited. The diffusion hyperintensity and lesion enhancement were significantly correlated (P <0.001). The sensitivity, specificity, PPV, NPV and accuracy were 100%, 67.9%, 32.3%, 100% and 72.1%, respectively. A hyperintense DWI finding does not necessarily overlap with contrast enhancement. There are many false positives, possibly representing other stages of lesion development. Although DWI may not replace CE T1WI imaging to demonstrate DIT due to the low PPV, it may serve as a screening MRI sequence where the use of GBCAs is a concern.