Automatic tumor image segmentation is challenging and difficult particularly in the nasopharynx area since the tumor is surrounded by various organs and image intensities of the tumor and normal tissue are in the similar level. Currently, radiologists manually draw the tumor regions on CT or MR images slide by slide. This is a tedious task and takes a long time. Automatic segmentation can reduce time and radiologists’ workload. In this research, region growing technique is used for the segmentation. With this technique, initial seed representing the tumor cell has to be generated.

Nearby seeds are then examined whether they are the tumor cells. In the original technique, seed is determined manually. Here, a technique for automatic seed generation based on probabilistic reasoning is proposed. This seed is then used for the region growing technique subsequently. CT images of patients who were treated at Ramathibodi hospital are tested and compared with the standard ground truth. The results showed that seed was generated with 87.5% accuracy. The segmented region provided a sensitivity of 0.88. However, this technique could be less accurately with the metastasis case.

3D tumor visualization is a useful tool for diagnosis and radiation treatment planning since physicians can exactly locate regions of tumor invasion. The tumor size and position can then be estimated correctly. Though there are some commercial software packages for the 3D visualization, they are costly. Thus they are not widely used in Thailand. In this research, software for 3D visualization of nasopharyngeal carcinoma from CT and MR images has been developed.

Thai physicians thus can use the developed software with the lower cost. This increases chances of patients in the rural area to receive better diagnosis and treatment. To create the 3D model, CT or MR images with identified tumor regions by radiologists are employed. The tumor regions gained from these images are reconstructed and registered onto the skull model subsequently. Tumor volume can be estimated. With this software, files from CT or MRI scans can be simply loaded into our software. Physicians can view and easily manipulate the constructed model.

The physicians thus can use these factors to indicate whether the cancer will recur in each patient prior to a thorough recheck. In this research, Neuro-Fuzzy based techniques will be used to predict the cancer recurrence. It combines the advantages of artificial neural networks and fuzzy logic. Thus it should provide better prediction performance. Additionally, using these techniques, we can extract knowledge pertaining risk of having cancer recurrence. The physicians can use the extracted knowledge to simply examine the cancer recurrence in each patient.

In the data collection, some patients might be lost to follow due to various reasons or they might have died because of other causes. These data are called the censored data and they cannot be used directly with the artificial intelligent techniques. Some modifications are thus required. From the previous research, the censored data were excluded from the data set. This resulted in biasing in the prediction. In this research, a technique that can deal with the censored data will be developed in order to provide effective prediction of the cancer recurrence.

Single Nucleotide Polymorphism or SNP is a genetic variation frequently found in a DNA sequence. It occurs when a single nucleotide differs in paired chromosomes. This genetic variation may affect to disease susceptibility or drug response in an individual. Identification of SNPs related to a disease is thus necessary in which it can enhance efficiencies in gene therapy. Additionally, these relevant SNPs can be used for predicting risk of having the disease. To detect such SNPs, a large number of case and control samples are required. In general, SNP microarray is used to detect the SNPs in an individual. However, it is costly particularly in large samples. Identification of SNPs from pooled DNA is thus applied. In this research, susceptibility loci of complex disease from pooled microarray data will be studied. Various feature extraction techniques will be examined.