116
A. Jlassi et al.
to allow this subdivision. Nevertheless, the visual inspection of CC structures
in MRI scans suffers from both inter- and intra-specialist variability. On the
one hand, the manual CC segmentation methods require strongly visual effort,
specialized training skill, and are time-consuming processes. On the other hand,
several geometrical methods for the CC parcellation have been proposed such as
Witelson and Hofer methods [12]. However, these methods cannot be satisfactorily validated due to the lack of qualitative parameters and reference standards.
Although all these difficulties, the development of an automatic CC parcellation
method is an inescapable need to ensure a reliable diagnosis. Such parcellation
is so independent from the operator skills and may be extended to other brain
structures parcellation. Thus, since there are no visible landmarks indicating
where the CC should be subdivided, the development of a fully automatic CC
parcellation method is highly challenging, even for specialists. To deal with this
issue, we propose to automatically parcel the CC within MRI images. By validating it, for the first time, on large and public datasets, the proposed method
records promising results. In fact, the contribution of this work is twofold:
– As best as we know, we adopt for the first time the superpixel segmentation
algorithm called Simple Linear Iterative Clustering (SLIC) for the CC parcellation [13]. Despite its simplicity, SLIC has been demonstrated to be effective
in various computer vision applications [14].
– The subdivision process of the proposed method is fully automatic and it is
the second study that proposed a non-geometric analysis for the CC parcels,
to the best of our knowledge [15]. Although it is based only on the MRI data
of each analyzed subject, with no parameter adjusting, the proposed method
proved quantitatively its superiority over state-of-the-art methods.
The rest of this paper is organized as follows. In Sect. 2, we briefly review
existing methods for the CC parcellation. Section 3 presents the proposed method
based on SLIC. Experimental results are discussed in Sect. 4. The last section
concludes the paper and points some directions for future work.
2 Related Work
Few CC parcellation methods were proposed. However, most of these methods
have not surmounted all the challenges encountered. In fact, the CC parcellation is a challenging task given that a normal shape of the CC might not clearly
highlight all parcels, what can increase the diagnosis complexity. In addition,
many internal abnormalities might include bumps which are hard to detect.
Existing CC parcellation methods can be divided into two main classes: geometric methods and non-geometric ones. On the one hand, since there are no real
or visible boundaries allowing the CC parcellation, several geometrical methods were presented to perform this task. Among these methods, two particular
ones are widely adopted. The first was proposed by Witelson and it is based
on postmortem connectivity analysis in primates and humans [16]. This method
divides the CC into five regions ranging from anterior dimension to the posterior
A. Jlassi et al.
to allow this subdivision. Nevertheless, the visual inspection of CC structures
in MRI scans suffers from both inter- and intra-specialist variability. On the
one hand, the manual CC segmentation methods require strongly visual effort,
specialized training skill, and are time-consuming processes. On the other hand,
several geometrical methods for the CC parcellation have been proposed such as
Witelson and Hofer methods [12]. However, these methods cannot be satisfactorily validated due to the lack of qualitative parameters and reference standards.
Although all these difficulties, the development of an automatic CC parcellation
method is an inescapable need to ensure a reliable diagnosis. Such parcellation
is so independent from the operator skills and may be extended to other brain
structures parcellation. Thus, since there are no visible landmarks indicating
where the CC should be subdivided, the development of a fully automatic CC
parcellation method is highly challenging, even for specialists. To deal with this
issue, we propose to automatically parcel the CC within MRI images. By validating it, for the first time, on large and public datasets, the proposed method
records promising results. In fact, the contribution of this work is twofold:
– As best as we know, we adopt for the first time the superpixel segmentation
algorithm called Simple Linear Iterative Clustering (SLIC) for the CC parcellation [13]. Despite its simplicity, SLIC has been demonstrated to be effective
in various computer vision applications [14].
– The subdivision process of the proposed method is fully automatic and it is
the second study that proposed a non-geometric analysis for the CC parcels,
to the best of our knowledge [15]. Although it is based only on the MRI data
of each analyzed subject, with no parameter adjusting, the proposed method
proved quantitatively its superiority over state-of-the-art methods.
The rest of this paper is organized as follows. In Sect. 2, we briefly review
existing methods for the CC parcellation. Section 3 presents the proposed method
based on SLIC. Experimental results are discussed in Sect. 4. The last section
concludes the paper and points some directions for future work.
2 Related Work
Few CC parcellation methods were proposed. However, most of these methods
have not surmounted all the challenges encountered. In fact, the CC parcellation is a challenging task given that a normal shape of the CC might not clearly
highlight all parcels, what can increase the diagnosis complexity. In addition,
many internal abnormalities might include bumps which are hard to detect.
Existing CC parcellation methods can be divided into two main classes: geometric methods and non-geometric ones. On the one hand, since there are no real
or visible boundaries allowing the CC parcellation, several geometrical methods were presented to perform this task. Among these methods, two particular
ones are widely adopted. The first was proposed by Witelson and it is based
on postmortem connectivity analysis in primates and humans [16]. This method
divides the CC into five regions ranging from anterior dimension to the posterior