Diagnostic accuracy of diffuse reflectance imaging for early detection of pre-malignant and malignant changes in the oral cavity: a feasibility study

Oro-pharyngeal cancer is a major component of the global cancer burden [1, 2]. Early detection of pre-malignant changes facilitates timely adoption of preventive measures and treatment strategies [3]. The potential beneficial effect of early detection and treatment may improve the survival rates in patients with cancer of the oral cavity [4]. The gold standard for diagnosis of oral cavity cancer is an invasive tissue biopsy and histopathology examination. However, detection of the potentially malignant site for biopsy is visually challenging even for experienced clinicians, which often leads to multiple biopsies and delay in diagnosis.

Toluidine blue staining [5] and direct fluorescence visualization [6, 7] have been used in clinical settings as an adjunctive visual tool to enhance the contrast between the clinical lesions and the adjacent normal oral tissue. Diffusely reflected (DR) white light spectra were also studied by various groups [8, 9] for tissue differentiation in oral cavity. Several multi-centric clinical studies established the effectiveness of optical spectroscopy techniques for non-invasive detection of oral malignancies with good diagnostic accuracies [10‐14]. However, they are point monitoring systems that analyse the tissue characteristics at a particular point in the entire area of an oral lesion.

Several spectroscopic imaging techniques that rely on tissue fluorescence have emerged recently for the detection of oral malignancies [6, 15‐17]. Our previous experience suggests that the diagnostic accuracies based on DR spectroscopy are superior to fluorescence spectroscopy in point monitoring systems. We therefore developed a multi-spectral diffuse reflectance imaging system for recording of DR images in vivo and report the diagnostic accuracies of this system in comparison to the gold standard tissue biopsy and histopathology examination.

All together 55 patients with oral lesions, such as erythroplakia, leukoplakia, non-healing ulcers, and mucosal growth, and 23 healthy volunteers as control were enrolled for the study. The study was carried out from April 2010 to December 2010. While the controls were significantly younger than cases, tobacco use and alcohol use were reported more frequently in cases than controls (Table 1). Out of the 55 patients examined using DR imaging, the pathology report showed that 6 lesions were epithelial hyperplasias with no dysplasia, 20 lesions were epithelial hyperplasia with different grades of dysplasia (mild, moderate and severe), and 29 lesions were malignant, which included well differentiated, moderately differentiated and poorly differentiated SCCs (Characteristics of the lesions are described in supplementary Additional file 1: Table S1). Epithelial hyperplasia with no dysplasia were included in the group of healthy and designated as normal/clinically healthy for analysis purpose. A total of 100 sites were examined from 23 healthy volunteers. Thus, the analysis was finally conducted by comparing 106- normal/clinically healthy sites, 20- pre-malignant (dysplasia) and 29- malignant (SCC) sites.

A non-healing ulcerated lesion in a 63 year old male patient on the right lateral border of the tongue that was present for more than two months is presented in Figure 2 along with its monochrome and false colored image ratios. While the bright yellow spot in the PCM indicates malignancy, the histo-pathological analysis of the biopsy sample confirms the diagnosis of well differentiated SCC (WDSCC).

By contrast, a lesion on the left lateral surface of the tongue in a 43 year old male patient, clinically diagnosed as speckled leukoplakia, histo-pathologically exhibits severe dysplastic features (Figure 3). The PCM image of this lesion shows red areas categorizing the lesion as pre-malignant.

While the median pixel value of the R545/R575 image ratio for normal/clinically healthy tissue was 0.87 (IQR = 0.82-0.94), they were 1.35 (IQR = 1.13-1.67) and 2.44 (IQR = 1.78-3.80) for pre-malignant and malignant lesion, respectively. Scatter plots based on R545/R575 image ratio are presented in Figure 4. While 1/29 malignant cases was misclassified as normal/clinically healthy, 3/106 normal/clinically healthy lesions were misclassified as malignant at a mean pixel intensity cut-off of 1.1 (97% sensitivity and specificity). Similarly a cut-off of 1.01 (Figure 4b) miss-classified 1/20 pre-malignant cases as normal/clinically healthy and 8/106 normal/clinically healthy cases as pre-malignant (sensitivity of 95% and specificity of 92%). While a cut-off at 1.66 (Figure 4c) misclassified 7/29 malignant lesions as pre-malignant and 4/20 pre-malignant lesions as malignant (sensitivity of 76% and specificity of 80%), a cut-off at 1.06 (Figure 4d) misclassified 4/49 cancerous or pre-malignant cases as normal/clinically healthy and 5/106 normal/clinically healthy cases as cancerous or pre-malignant (sensitivity of 92% and specificity of 95%). Discordant findings of histopathology and DR imaging results are presented in Table 2 with their detailed histopathology reports. Diagnostic accuracies and positive and negative predictive values are presented in Table 3. Area under the ROC curves (Figure 5a-d) show the discriminatory capacity of the image ratio to differentiate malignant from normal/clinically healthy [AUC = 0.99 (95% CI: 0.99-1.00)], pre-malignant from normal/clinically healthy [AUC = 0.94 (95% CI: 0.86-1.00)], malignant from pre-malignant [AUC = 0.84 (95% CI: 0.73-0.95)] and pre-malignant and malignant from normal/clinically healthy [AUC = 0.97 (95% CI: 0.94-1.00)] lesions.

The present study examines the diagnostic accuracy of a spectral imaging system based on the principles of diffuse reflectance in discriminating healthy oral tissue from pre-malignant and malignant tissues. The diagnostic accuracy obtained in this study is superior to other imaging systems for detection of malignant changes in the oral cavity.

When light interacts with biological tissue a small portion of it is absorbed or transmitted while the rest undergoes multiple elastic scattering due to heterogeneity in the refractive index of the tissue components and gets diffusely reflected. Often, a portion of the impinging radiation is reflected from the surface when the roughness of the boundary is small in comparison with the wavelength of light. The portion that penetrates the sample gets scattered at a larger number of points in its path due to uneven, broken or bumpy boundary surfaces, where the coarseness is of the same order of magnitude as the wavelength. During tissue transformation healthy tissue undergoes morphometric and cytologic changes such as increase in epithelial thickness, nuclear size, nuclear to cytoplasmic ratio, changes in the chromatin texture and collagen content, and angiogenesis [20‐22]. These changes modify the diffusely reflected component of the incoming radiation. We hypothesized that the diffuse reflectance intensity decreases during the transformation of tissue from healthy to dysplastic due to morphologic, cytologic and vascular changes associated with transformation. While our previous studies supported this hypothesis and differentiate pre-malignant and malignant tissue with relatively good sensitivity and specificity, the present study findings further strengthens this relationship and enables identification of areas with the most malignant potential in an oral lesion.

Although the biological mechanisms associated with changes in DR of different tissue types are not clear, one potential possibility is associated with changes in production of haemoglobin. In malignant tissues, the haeme synthesis is disturbed due to the reduced activity of the ferrochelatase enzyme [23] that results in lower haemoglobin production and correspondingly lower absorption at 545 and 575 nm of the oxygenated haemoglobin spectra. A reduction in oxygenated haemoglobin increases the DR ratio of R545/R575. Conversely, during inflammatory conditions there is an increase in haeme production, which leads to an enhancement in the oxygenated haemoglobin and concomitant decrease in the DR ratio of R545/R575 (Figure 6).

The diagnostic accuracies obtained in our study are superior to the diagnostic accuracies obtained in studies using direct tissue fluorescence visualization [6, 16, 17]. Unlike the present study, the discriminative capacity was tested in a homogenous group of malignant lesions consisting of mainly severe dysplastic tissues in all other studies. The high diagnostic accuracy obtained in our study underlines the potential use of this method in routine clinical practice. However, we could not perform site-specific analyses due to small number of cases in each anatomical location. The diagnostic accuracies presented in our study are therefore averaged for all types of lesions of the oral cavity and it would have been better if site-specific lesion classifications were possible.

The basis for our DR imaging method arises from the differences in the spectra obtained from the normal and diseased tissue due to the multiple physiological changes associated with tissue transformation from healthy to pre-malignant/dysplastic and malignant. The imaging method has the advantage of non-invasively scanning the entire lesion and its surrounding areas in real-time, and categorize oral lesions into normal/clinically healthy, pre-malignant and malignant tissue. Furthermore, it efficiently delineates the boundaries of neoplastic changes and locates the site with most malignant potential for a biopsy, thereby avoiding unnecessary repeated biopsies and delay in diagnosis. Imaging the entire region may also help the surgeons to identify the margins of the lesion that cannot be easily visualized by the naked eye during surgical interventions. Applications of digital image processing techniques may further enhance our ability to objectively identify and delineate the peripheral extent of neoplastic lesions.

Since clinical diagnosis based on DR imaging is possible in near real-time, there is practically no waiting period for the patient. The method is relatively cheap and can be implemented in all clinical settings with minimal training. Training the non-physician health workers in the imaging technique and screening the patients for tissue biopsy may further reduce the cost. However, the cost-effectiveness of mass screening of oral lesions using the imaging technique needs to be evaluated in different settings before its wider adaptation. Furthermore, the multi-spectral DR imaging technique presented in this paper has the potential to be used as an adjunct to colposcopy in the screening of cervical pre-cancers and in the identification the most malignant site for biopsy. Further studies in a larger population may help us to develop better classification algorithms for discriminating dysplastic lesions as mild, moderate and severe, and SCC lesions as ‘well differentiated’, ‘moderately differentiated’ and ‘poorly differentiated’ tissue categories.

Diffuse reflectance spectral imaging technique efficiently differentiates healthy tissue from pre-malignant and malignant tissue in the oral cavity. The imaging system developed for this study can be used as an adjunctive visual tool to enhance the contrast between dysplastic tissue and normal tissue in oral cavity lesions in clinical settings.