La reconstrucción de lo deteriorado en Bogotá

Originally we limited the scope of the “Tumor-cut” algorithm to the contrast en- hanced gross tumor volumes, which corresponds to the tumor core of the high grade tumors in Table 3.2. The comparison of the performance with the other partici- pating algorithms shows that the “Tumor-cut” performs superior to others for the segmentation of the tumor core of high grade glioma cases, whereas perform similar for other categories, except edema segmentation of the low grade tumors, which is usually not of interest in low grade gliomas.

For the 4-label challenge results given in 3.4, “Tumor-cut” algorithm performed comparable to the state-of-the-art techniques participated in the challenge. Because, the “Tumor-cut” method is not based on learning on the training dataset, the results obtained for both training and testing datasets are similar as expected. However, the low ranking in Table 3.5 is possibly due to the learning-based biased performance evaluation of the other algorithms.

Disadvantages of the interactive segmentation algorithms in general can be noted as: increased processing time and the possibility to miss some in case of multiple tumors, difficulty in batch processing of patient databases and intra/inter operator variability.

We observed that in one case only, we segmented an abnormal structure, which was not labeled as tumor by the experts. Although, this resulted a zero overlap score for the particular case, in fact, to allow user to choose what to segment is an advantage of the semi-automatic approach. In general, the T2 results did not provide useful information, as only a small portion of the tumors consist of the non- enhancing region and the segmentation results were not accurate due to the low contrast between tumor core and edema. The approach of “Tumor-cut” method was to apply the original algorithm independently to each modality. A combined algorithm that considers the multidimensional information from all available modal- ities have the potential to improve the results obtained.

Chapter 4

Potential Tumor Response

Criteria based on the Invariants of

the Finite Strain Tensor

1

4.1

Tumor Follow-up

The work on developing a common language, in order to evaluate the therapy re- sponses of the tumors in an objective and consistent manner, was accelerated after mid 1970’s. For this purpose, the World Health Organization (WHO) criteria titled ”Reporting results of cancer treatment” which is published by World Health Orga- nization (WHO) in 1979 [54] and by Miller et.al. in 1981 [55] were widely used as a standard in the studies. WHO criteria proposes to calculate the total tumor load to determine the treatment response by multiplying the maximum diameter in any of the three orthogonal planes (axial, sagittal or coronal) and the maximum diameter in the perpendicular plane for each tumor and summing up over measurable lesions (2 dimensional). The measurement of the tumor load should be done at the begin- ning of the treatment and with equal time interval follow-ups by the same method. Whenever the accurate measurement in 2 dimensions is not possible, only the the single longest diameter of the tumor would be reported. Treatment responses for the measurable diseases are classified under 4 categories:

1_{The work presented in this chapter appeared in the European Society of Magnetic Resonance}

• Complete Response (CR): The disappearance of all known disease.

• Partial Response (PR): No new lesion, no progression of any lesion and at least 50% decrease in total tumor load.

• Progressive Disease (PD): A 25% or more increase in the size of any measurable lesions or the appearance of new lesions.

• No change (NC): No response or progression.

Complete and partial responses should be verified by a second measurement following the observation. [47].

In 1990, after the publication of the application of WHO criteria on brain tumors by Macdonald et.al., MacDonald criteria became a standart used on brain tumor studies [10, 11].

However, summing up the multiplication of bidimensional measurements, needs excessive workload and the results are error-prone and causes some problems about the usage of the WHO criteria [56, 57, 58]:

1. The usage of the tumor size changes in response evaluation varies between research groups.

2. The number of lesions and the minimum lesion sizes noted shows variations. 3. ”Progessive Disease (PD)” is defined on a single lesion by some groups, while

calculated on total tumor load by some other groups.

4. The technological development of the measurement devices (Computed To- mography and Magnetic Resonance Imaging), causes confusions on how to use 3-dimensional measurements on tumor response studies.

Addressing those problems, in the mid 1990’s, a study is initiated to develop a new set of criteria by ”The European Organization for Research and Treatment of Cancer (EORTC)”, ”The National Cancer Institute (NCI) of the United States”, and ”The National Cancer Institute of Canada, Clinical Trials Group” and published in the year 2000, under the name Response Evaluation Criteria in Solid Tumors (RECIST) [59]. In this widely used RECIST tumor response criteria, the maximum

diameter of the tumor measured in any plan is used for tumor follow-up. Comparison studies revealed that the overall results obtained by unidimensional measurements of a single maximum diameter (RECIST) are equivalent to the bidimensional WHO criteria [56, 59, 58, 60].

In this chapter, as well as the global criteria such as diameter and volume mea- surements, the potential of the local criteria on assessment of the tumor evolution is studied and explained in the following section.