Page 106 - The-5th-MCAIT2021-eProceeding

4. Proposed Experimental Evaluation

This research intends to use accuracy (Mahayuddin & Saif , 2020; Saif et al. 2015;Schawkat et al., 2020),
sensitivity (True Positive Rate) (Saif &Mahayuddin, 2020; Saif et al., 2015 ; Ide et al. ,2020) and specificity
(True Negative Rate) (Saif et al., 2014 ;Beckett et al. ,2020) as primary performance metrics for segmenting
and detecting tumor region from whole brain MRI image. All three metrics can be calculated using equation
(1), (2) and (3).
Sensitivity = TP / (TP + FN) (1)
Specificity = TN / (TN + FP) (2)
Accuracy = (TN +TP) / (TN +TP + FN + FP) (3)

True positive (TP) or tumor pixels marked as tumor which can be measured by checking how many of the
ground truth tumorous pixels are marked as tumor pixel in the segmented tumor image. False positive (FP) or
non-tumor pixels marked as tumor which can be measured by checking how many of the background pixels
(non-tumor pixels) are marked as tumor pixels in the segmented tumor image. True negative (TN) or non-tumor
pixels marked as non-tumor which can be measured by checking how many of the background pixels of the
ground truth image are marked as background pixels in the segmented tumor image. False negative (FN) or non-
tumor pixels marked as tumor. which can be measured by checking how many of the tumor pixels in the ground
truth image are marked as background pixels in the segmented image. Sensitivity or true positive rate indicates
the proportion of pixels of tumor region in MRI segmented correctly as tumor pixel. Specificity or true negative
rate indicates the proportion of non-tumor pixels that are correctly segmented or marked by segmentation mask
as non-tumor pixel. Accuracy is the rate of how much correctly the whole tumor is segmented, that is tumor
marked as tumor and non-tumor marked as non-tumor together. For validation purpose, this research intends
datasets from figshare (Divya et al. , 2020 ; Deepak and Ameer, 2020; Bulla et al., 2020) which contains 3064
T1-weighted contrast-enhanced images from 233 patients with three kinds of brain tumor: meningioma (708
slices), glioma (1426 slices), and pituitary tumor (930 slices).
This research aims to use convolutional neural network for classification purpose after preprocessing step.
In this context, MRI images are normally full of noises. Improved pre-processing of image data can be reduced
as noise free sand prepared an image for next steps such as segmentation. In the case of segmentation, existing
research methods reached over 70% to 80% accuracy, but for ensuring a safe surgery and other treatment
method, tumor must be detected with 100% or as close to 100%. This research aims to establish an efficient
segmentation method that can achieve an accuracy of almost 100% with as less complex computation method
as possible.

5. Conclusion

Research on cancer cell and tumor detection and segmentation from MRI images is a valuable working area
of modern computer vision research domain. Magnetic Resonance Imaging (MRI) is a medical imaging
technique that is being used for cancer and tumor detection mostly nowadays. In this context, segmentation of
medical imagery is a very challenging issue as MR images are full of many important information about patient’s
health and it is somewhat noisy to watch. This research proposed a conceptual framework to segment cancer
cell from brain MRI image and later convolution network is considered for further processing. This research
aims to use T1 MRI image dataset of brain tumor containing meningioma tumors and brain tumor segmentation
(BraTS) challenge dataset of brain tumor. Later, experimental evaluation strategy is proposed to validate the
proposed conceptual framework which is expected to contribute immensely in medical image processing
domain. In the near future, proposed conceptual framework will be investigated extensively to establish the
proposed methodology.

E- Proceedings of The 5th International Multi-Conference on Artificial Intelligence Technology (MCAIT 2021) [93]
Artificial Intelligence in the 4th Industrial Revolution