Automated Histological Image Analysis
Digital histological analysis has grown in popularity recently as increases in computational resources have allowed for fast and inexpensive processing of high-resolution images. Concomitant with this shift has been a tendency towards providing quantitative data in addition to traditionally qualitative descriptions of histological analyses.
Thoracic injuries are exceedingly common in blunt trauma patients and are associated with significant morbidity and mortality. PC is a common injury following blunt chest trauma, effecting up to 25% of patients sustaining blunt chest trauma and between 10 and 17% of all trauma admissions. The sequela of pulmonary contusion varies widely, ranging from mild dyspnea to prolonged mechanical ventilation, infection, and Acute Respiratory Distress Syndrome (ARDS).
Contusion volume measured at admission has been shown to be an independent predictor for subsequent development of ARDS, with the risk of ARDS increasing sharply with PC in excess of 20% by volume. The wider availability of such thresholds is hindered by the fact that quantitating PC remains a time intensive and difficult task.
|
|
|
Top, Image of H&E stained lung section showing areas of healthy and pathologic lung parenchyma. Middle, Pathologist identified contusion. Bottom, Contusion identified through the use of the automated algorithm. |
The purpose of the present study is to demonstrate the effectiveness of an algorithm developed in our lab to differentiate regions of pathologic lung tissue from healthy lung tissue in a set of histological samples. The results of the algorithm can be used to assess the degree of contusion in future studies.
Hemoatoxylin and eosin-stained slides of contused parenchyma were digitally imaged. Images were imported in Matlab as unsigned, 8-bit, RGB images for preprocessing and analysis. Each image was preprocessed to clearly distinguish background from lung and to translate each image so that all images in the stack were aligned along a common centroid. Various morphological techniques are then used to accentuate and identify areas of contusion.
For algorithm validation, an experienced pathologist also quantitated the amount of contusion independently using a standardized procedure. The image analysis algorithm is currently being honed to best match the pathologist regions of contusion identified by the pathologist.