Modify figure for vertical and horizontal traversal

joss/paper.md

@@ -42,7 +42,7 @@ Medical imaging data is shared in various file formats (such as DICOM, NIfTI, An
 
 Prior works [@Yarkoni2019;@brett_2023_7795644] have acknowledged the complexity of working with large scale, heterogeneous datasets. However, there has been limited emphasis on providing a unified interface for accessing various subjects and sequences without reformatting, that goes beyond one file format and dataset structure. It is important to note that in the normal course of a typical research project that involves many open datasets, importing, reformatting and validation of datasets typically occurs years after the data acquisition process has been completed. Relying on a specialized tool designed exclusively for a specific format is not practical. A unified interface helps decouple the analyses from the constraints and idiosyncrasies of various formats. Besides being user-friendly for accessing multiple formats, MRdataset enables direct access to sessions/subjects to facilitate sequence-level and subject-level analyses. MRdataset captures the broader perspective of reading the dataset as a whole and provides intuitive methods for horizontal traversal (across subjects for a given sequence) and vertical traversal (across multiple sequences for a given subject) as shown in \autoref{fig:schematic}. \autoref{fig:loops} shows an example of a script accessing subjects and sequences from a DICOM dataset.
 
-![ Illustration of horizontal and vertical traversal for accessing subjects/sequences of a medical imaging dataset\label{fig:schematic} ](Slide8.jpg)
+![ Illustration of horizontal and vertical traversal for accessing subjects/sequences of a medical imaging dataset\label{fig:schematic} ](Slide8.png)
 
 In addition to the differences in study design and imaging formats, imaging scans also exhibit differences in MR physics parameters due to scanner vendors (such as Siemens, GE, and Philips) and hardware/software version updates. MRdataset accommodates such variations through custom classes for each acquisition parameter (via the protocol library we developed). These classes encapsulate contextual information, including their physical units, value range and their level of criticality to subsequent analyses. Utilizing specific classes for each acquisition parameter provides a rich, native and unambiguous representation of these parameters. They would also allow easy extensibility to accommodate distinct study designs and requirements. For example, we capture multiple echo times and flip angles in an array-like fashion although they are scattered in different DICOM slices. Subsequently, they are stored in a manner tailored to align with the needs of researchers.