Estimating Bacterial and Cellular Load in FCFM Imaging

We address the task of estimating bacterial and cellular load in the human distal lung with fibered confocal fluorescence microscopy (FCFM). In pulmonary FCFM some cells can display autofluorescence, and they appear as disc like objects in the FCFM images, whereas bacteria, although not autofluorescent, appear as bright blinking dots when exposed to a targeted smartprobe. Estimating bacterial and cellular load becomes a challenging task due to the presence of background from autofluorescent human lung tissues, i.e., elastin, and imaging artifacts from motion etc. We create a database of annotated images for both these tasks where bacteria and cells were annotated, and use these databases for supervised learning. We extract image patches around each pixel as features, and train a classifier to predict if a bacterium or cell is present at that pixel. We apply our approach on two datasets for detecting bacteria and cells respectively. For the bacteria dataset, we show that the estimated bacterial load increases after introducing the targeted smartprobe in the presence of bacteria. For the cell dataset, we show that the estimated cellular load agrees with a clinician’s assessment.

[1]  Andrew Zisserman,et al.  Interactive Object Counting , 2014, ECCV.

[2]  R. Rodríguez-Roisín,et al.  Validation of different techniques for the diagnosis of ventilator-associated pneumonia. Comparison with immediate postmortem pulmonary biopsy. , 1994, American journal of respiratory and critical care medicine.

[3]  Christopher K. I. Williams,et al.  Localisation microscopy with quantum dots using non-negative matrix factorisation. , 2014, Optics express.

[4]  A. Akram,et al.  A labelled-ubiquicidin antimicrobial peptide for immediate in situ optical detection of live bacteria in human alveolar lung tissue† †Electronic supplementary information (ESI) available: Experimental details and Fig. S1–S5. See DOI: 10.1039/c5sc00960j , 2015, Chemical science.

[5]  A. Torres,et al.  Bronchoscopic BAL in the diagnosis of ventilator-associated pneumonia. , 2000, Chest.

[6]  Sohan Seth,et al.  Assessing the utility of autofluorescence-based pulmonary optical endomicroscopy to predict the malignant potential of solitary pulmonary nodules in humans , 2016, Scientific Reports.

[7]  G. Bourg-Heckly,et al.  Human in vivo fluorescence microimaging of the alveolar ducts and sacs during bronchoscopy , 2009, European Respiratory Journal.

[8]  Marc Decramer,et al.  Acute lung allograft rejection: diagnostic role of probe-based confocal laser endomicroscopy of the respiratory tract. , 2014, The Journal of heart and lung transplantation : the official publication of the International Society for Heart Transplantation.

[9]  Luc Van Gool,et al.  Efficient Non-Maximum Suppression , 2006, 18th International Conference on Pattern Recognition (ICPR'06).

[10]  Andrew Zisserman,et al.  Counting in the Wild , 2016, ECCV.

[11]  Sohan Seth,et al.  Estimating Bacterial Load in FCFM Imaging , 2017, MIUA.

[12]  Stephen McLaughlin,et al.  Automated Detection of Uninformative Frames in Pulmonary Optical Endomicroscopy , 2017, IEEE Transactions on Biomedical Engineering.

[13]  V. Ranieri,et al.  Diagnosis of ventilator-associated pneumonia: a systematic review of the literature , 2008, Critical care.

[14]  Melih Kandemir,et al.  Gaussian Process Density Counting from Weak Supervision , 2016, ECCV.

[15]  Sergei Vassilvitskii,et al.  k-means++: the advantages of careful seeding , 2007, SODA '07.