Artificial Intelligence Applications in Oral Cancer and Oral Dysplasia

Abstract

Oral squamous cell carcinoma (OSCC) is a highly unpredictable disease with devastating mortality rates that have not changed over the past decades, in the face of advancements in treatments and biomarkers, which have improved survival for other cancers. Delays in diagnosis are frequent, leading to more disfiguring treatments and poor outcomes for patients. The clinical challenge lies in identifying those patients at the highest risk of developing OSCC. Oral epithelial dysplasia (OED) is a precursor of OSCC with highly variable behavior across patients. There is no reliable clinical, pathological, histological, or molecular biomarker to determine individual risk in OED patients. Similarly, there are no robust biomarkers to predict treatment outcomes or mortality in OSCC patients. This review aims to highlight advancements in artificial intelligence (AI)-based methods to develop predictive biomarkers of OED transformation to OSCC or predictive biomarkers of OSCC mortality and treatment response. Biomarkers such as S100A7 demonstrate promising appraisal for the risk of malignant transformation of OED. Machine learning-enhanced multiplex immunohistochemistry workflows examine immune cell patterns and organization within the tumor immune microenvironment to generate outcome predictions in immunotherapy. Deep learning (DL) is an AI-based method using an extended neural network or related architecture with multiple "hidden" layers of simulated neurons to combine simple visual features into complex patterns. DL-based digital pathology is currently being developed to assess OED and OSCC outcomes. The integration of machine learning in epigenomics aims to examine the epigenetic modification of diseases and improve our ability to detect, classify, and predict outcomes associated with epigenetic marks. Collectively, these tools showcase promising advancements in discovery and technology, which may provide a potential solution to addressing the current limitations in predicting OED transformation and OSCC behavior, both of which are clinical challenges that must be addressed in order to improve OSCC survival.

Keywords: digital pathomics; epigenomics; multiplex immunohistochemistry; oral epithelial dysplasia; oral squamous cell carcinoma.

FIG. 1.

Clinical presentation of oral epithelial dysplasia. (A) Leukoplakia of the lateral tongue. (B) Erythroplakia of the dorsolateral tongue. (C) Erythroleukoplakia of the lateral tongue.

FIG. 2.

Photomicrographs of oral dysplasia and squamous cell carcinoma. (H&E stain; ×100) (A) Normal stratified squamous epithelium with no evidence of dysplastic changes. (B) Mild dysplasia: epithelium exhibiting hyperchromatic and slightly pleomorphic nuclei in the lower one-third of epithelium. Additionally, basal cell disorganization is also seen (black solid arrows). (C) Moderate dysplasia: epithelium exhibiting cellular pleomorphism, hyperchromasia, and loss of polarity. Atypia is seen affecting about half the thickness of the epithelium (black solid arrow). Marked hyperkeratosis is also seen (black dotted arrow). (D) Severe dysplasia: epithelium exhibiting marked pleomorphism, hyperchromasia, variably sized nuclei and cells, and scattered mitotic figures (black dotted arrow). Atypia is seen affecting most of the thickness of the epithelium (green solid arrow). Hyperkeratosis is also seen (black solid arrow). (E) Squamous cell carcinoma: well differentiated: islands of malignant squamous epithelial cells invading the fibrous connective tissue (black dotted arrow). Chronic inflammation is seen interspersed with the tumor islands (blue dotted arrow). Keratin pearls are also observed (black solid arrow).

FIG. 3.

S100A7 biomarker used in oral dysplasia risk prediction. Straticyte^TM creates a digital image and identifies regions of interest(s) using S100A7 staining. The program designates these areas with red = S100A7 negative area, maroon/purple = S100A7 biomarker positive area, green = S100A7 biomarker negative nuclei, blue = S100A7 biomarker positive nuclei, and teal = S100A7 positive areas in a stepwise work up, and only areas inside are used for risk calculation. Each area is then weighted and factors into the risk transformation calculation. To simplify clinical interpretation, samples are graded as either “low risk” or “elevated risk,” with the elevated risk having two subcategories. Low risk has a cancer transformation risk of 21% (95% confidence interval [CI]: 12–35%). Elevated risk category 1 has a transformation risk of 49% (95% CI 39–60%). Elevated risk category 2 has a transformation risk of 76% (95% CI: 61–89%). Kaplan–Meier survival curve red line shows the probability of cancer transformation each year, black line shows the upper 95% CI:, and blue line shows the lower 95% CI. Scale bar bottom right = 100 µm. Images provided by Jason Hwang, PhD (Proteocyte).

FIG. 4.

General pipeline for deep learning-based pathology. (A) H&E histology slides are digitally scanned into WSIs. (B) Tiles are extracted from annotated tumor regions. (C) Tiles are preprocessed through rotations and stain normalization and then labeled with the prediction of interest (e.g., HPV status in HNSCC). (D) The dataset is split for training and testing the DL neural network model. (E) After training, the model’s prediction performance on the test set is evaluated by metrics such as AUROC. The model may also be tested on an external, never-before-seen dataset of WSIs. AUROC, area under the receiver operating characteristic; HNSCC, head and neck squamous cell carcinomas; HPV, human papillomavirus; WSIs, whole slide images.

FIG. 5.

Summary diagram of the different artificial intelligence methods used in oral dysplasia and oral cancer biomarker research.