Pattern recognition methods for early detection of precancerous lesions in South Asia
Oral cancer is the most common cancer and constitutes a major health problem in the developing countries. Each year 270,000 people globally die from oral cancer. In contrast to the western countries, the incidence of oral cancer is considerably higher in South and Southeast- Asian developing countries. In India, ca 80,000 new oral cancer cases are annually reported. This makes oral cancer one of the most prevalent cancers (35% of all cancers causing a high death toll each year in India). Oral cancer is therefore a major health problem in India. In fact, in some parts of India, oral cancer accounts for more than 50% of all cancer.
Underlying clinical problem
One of the most important factors determining the clinical outcome of the disease is to detect the tumour in time, before it has spread to regional lymph nodes or generated metastases. The etiology of oral cancer is unknown, but important predisposing factors for oral cancer are related to the habit of tobacco quid chewing, smoking and excessive consumption of alcohol.
Although the carcinogenesis is believed to begin on molecular level (DNA damage) there are no cost-effective methods for screening potential individuals for DNA damage. The common situation is that the patient arrives at the clinic when the lesion is visually manifested and seen on the clinical level. Analysis is performed from clinical features to cellular level and further to molecular level; the latter is not yet common at clinical practice. The diagnosis of oral lesions regarded as clinically pre-cancerous lesions is challenging because the clinical appearance alone (with naked eyes) is not diagnostic. Moreover, the similar situation might exist between molecular and cellular level. At present, a biopsy and histopathological result is used to confirm the diagnosis, the setback is that the histological interpretation is to some extent subjective, since it varies among pathologists. Several studies have shown great interexaminer and intraexaminer variability in the assessment of the presence or absence and the grade of oral epithelial dysplasia in the histological samples, which justifies the need for the computerized decision support in oral histopathology.
Clinical benefits with concept
The project will contribute to improved detection and diagnosis of potentially precancerous oral lesions. This may lead to an early detection of precancerous lesions and in the long term to lower mortality of oral cancer rates. It will capture the pathologist’s knowledge into an organized and computerized system, using state-of the-art algorithms.
The project will fulfill the WHO-organization research policy involving the strengthening of health research systems. It contributes to health system development and health improvement particularly in poorer countries by: (1) the dissemination and translation of valuable knowledge or research, (2) the promotion and implementation of high quality health research evidence.
Medical procedure
The main task to be solved is the design of classification system with respect to the clinical diagnosis such as Oral Leukoplakia, Oral Erythroplakia, Oral Submucous Fibrosis. These lesions represent the most of the high-risk premalignant oral lesions. Another main task is the design of the classification with respect to different oral epithelial dysplasia classification systems. The three classification schemes: (I) oral epithelial dysplasia scoring system (II) squamous intraepithelial neoplasia and (III) Ljubljana classification, are planned to be evaluated and computerized.
Project description and objectives
The objective of this project is to develop a computer-aided diagnosis system for mucosal disorders in the oral cavity, using histology images as input source, to support the histopathological evaluation of the biopsies. In particular we will investigate oral lesions clinically diagnosed as leukoplakia, erythroplakia and oral submucous fibrosis. These lesions are regarded as most likely to develop into cancer and it is important to discover them at early stage.
In the preliminary stage we have delevoped a classification system that distinguish between normal oral mucosa from oral lichen planus and oral submucous fibrosis. The classification is based on image features extracted from histology images of the investigated cases. The system accuracy is ca 95%.
Project leader and R&D office
Artur Chodorowski, Chalmers University of Technology
Prof. Vinay K Hazarey, Dept of Oral Pathology, Govt. Dental College & Hospital, Nagpur, India
Dr. Sunil Kumar Kothawar, Dept of Oral Pathology, Sharad Pawar Dental College and Hospital, Sawangi, Wardha, India
