Towards a non-invasive diagnostic aid for abdominal adhesions using dynamic MRI and image processing

This work presents a strategy for detection of abdominal adhesions based on cine-MRI data, image processing and the production of a ‘sheargram’. Abdominal adhesions are a common complication of abdominal surgery and can cause serious morbidity. Diagnosis is difficult and often one of exclusion. A conclusive diagnosis typically requires laparoscopic explorative surgery, which itself may cause further adhesions. A non-invasive means of diagnosis is preferred and likely to aid patient management. Cine-MRI can capture the motion of the abdominal structures during respiration and has shown promise for adhesion detection. However, such images are difficult and time consuming to interpret. A previous PhD considered augmenting cine-MRI by quantifying movement for detection of gross adhesive pathology. This thesis presents a refined image processing approach aimed at detection of more subtle adhesions to the abdominal wall.

In the absence of adhesive pathology, the abdominal contents (bowels, kidneys, liver) slide smoothly against the perimeter of the abdominal cavity – a process termed visceral slide. An adhesion is expected to produce a localised resistance that inhibits smooth visceral sliding. In this PhD, development of a 2D technique to quantify sliding around the perimeter of the abdominal cavity (with particular emphasis on the abdominal wall) sought to highlight regions of reduced sliding. Segmentation and image registration were employed to quantify movement and shear, the latter used as an analogue for sliding. The magnitude of shear over all frames in the dynamic MR image sequence was extracted and displayed as a colour plot over the MR image for anatomical context. This final output is termed a ‘sheargram’. Suitability of the technique for diagnosis was assessed through a series of experimental tests and correlation with clinical data. The latter involved a retrospective pilot study incorporating data from 52 patients scanned for suspected adhesions. A total of 141 slices were processed and reported.

The validation experiments confirmed the technique had the attributes to accurately and reproducibly report sliding and demonstrated proof of concept for detection of adhered regions. The pilot study confirmed the sheargram matched expert clinical judgement in the vast majority of cases (>84%) and detected >93% of all adhesions. However, the investigation also highlighted limitations, principally structures moving out of the imaging plane creates a fundamental problem and requires a 3D imaging solution. In conclusion, the work has produced encouraging results and merits further development.