Niranjan Joshi

The segmentation of colorectal MRI images.

One of the key criteria that informs patient management decisions for colorectal cancer is the extent of the shortest distance from the edge of the primary tumour to the edge of the mesorectum, also referred to as circumferential resection margin (CRM). This region is resected during surgery. The CRM is difficult for clinicians to measure accurately, particularly from 2D slice data. We present a method for automatically calculating and visualising the CRM distances in colorectal cancer MR images. We use local phase of the monogenic signal calculated from the MR image intensities to find edge and ridge features within the data. A non-parametric mixture model is then used to describe image intensity values within level set framework in order to segment the mesorectal fascia and the corresponding tumour and lymph nodes, as distinct regions. This segmentation is used to provide an automatic analysis of the shortest distance resection margin, and we show that this is consistent with that of the clinically accepted MERCURY method. We use the segmentation to provide a 3D visualisation of where the resection margin is smallest. Finally, we reconstruct a 3D map of the segmented anatomy. Both the visualisation methods provide a useful tool to aid surgeons in their treatment planning.

Simplified Computation for Nonparametric Windows Method of Probability Density Function Estimation

Recently, Kadir and Brady proposed a method for estimating probability density functions (PDFs) for digital signals which they call the Nonparametric (NP) Windows method. The method involves constructing a continuous space representation of the discrete space and sampled signal by using a suitable interpolation method. NP Windows requires only a small number of observed signal samples to estimate the PDF and is completely data driven. In this short paper, we first develop analytical formulae to obtain the NP Windows PDF estimates for 1D, 2D, and 3D signals, for different interpolation methods. We then show that the original procedure to calculate the PDF estimate can be significantly simplified and made computationally more efficient by a judicious choice of the frame of reference. We have also outlined specific algorithmic details of the procedures enabling quick implementation. Our reformulation of the original concept has directly demonstrated a close link between the NP Windows method and the Kernel Density Estimator.

Random walk-based automated segmentation for the prognosis of malignant pleural mesothelioma

In this paper we apply the random walk-based segmentation method to mesothelioma CT image datasets, aiming to establish an automatic segmentation routine that can provide volumetric assessments for monitoring progression of the disease and its treatments. We have validated the applicability of this method to our image data through a series of experimental trials, and demonstrated the superior performance and benefits of random walk compared to other segmentation algorithms such as level sets.

Estimating the mesorectal fascia in MRI

Apart from chemoradiotherapy, surgery by total mesorectal resection is currently the only curative therapy for colorectal cancer. However, this often has a poor outcome, especially if there are affected lymph nodes too close to the resection boundary. The circumferential resection margin (CRM) is defined as the shortest distance from an affected region to the mesorectal fascia (MF), and should be at least 1 mm. However, this 3D distance is normally estimated in 2D (from image slices) and takes no account of uncertainty of the position of the MF. We describe a system able to estimate the location of the MF with a measure at each point along it of the uncertainty in location, and which then estimates the CRM in three dimensions. The MF localisation algorithm combines anatomical knowledge with a level set method based on: a non-parametric representation of the distribution of intensities, and the use of the monogenic signal to detect portions of the boundary.

Non-parametric Mixture Model Based Evolution of Level Sets

We present a novel region based level set algorithm. We first model the image histogram with non-parametric mixture of probability density functions(PDFs). The individual densities are estimated using a recently proposed PDF estimation method which relies on a continuous representation of the discrete signals. Prior probabilities are calculated using an inequality constrained least squares method. The log ratio of the posterior probabilities is used to drive the level set evolution. We also take into account the image artifact called the partial volume effect, which is quite important in medical image analysis. Results are presented on natural as well as medical two dimensional images. Visual inspection of our results show the effectiveness of the proposed algorithm

Comparison of classifier performance for information fusion in automated Diabetic Retinopathy screening

Diabetic Retinopathy (DR) is a vascular disorder affecting the retina due to prolonged Diabetes. It can lead to sudden vision loss in advanced stages. Screening and routine monitoring is the most effective way of avoiding vision loss due to DR. Abramoff et al.[1] developed and evaluated an automated DR screening system. One of the most important parts of this system, the information fusion module, combines information obtained from different images and various image properties. Niemeijer et al. [2] compared several methods for DR information fusion and concluded that k-Nearest Neighbour (kNN) provided the best performance for their system. The aim of this work was to compare performance of the Random Forest (RF) classifier with that of the kNN classifier for DR information fusion. We performed experiments on a dataset containing images from 10303 eye examinations. Additionally we also compared performance of the two classifiers in an important sub-problem of DR screening - red lesion detection. In both the experiments, the RF classifier showed significantly better performance.

Tramline and NP windows estimation for enhanced unsupervised retinal vessel segmentation

This paper presents a novel unsupervised vascular segmentation algorithm which is applied to retinal fundus images, however could be generalised to any two-dimensional vascular image. The algorithm presents a new fully automatic framework for vessel segmentation and comprises the following features: novel application of the NPWindows method for intensity distribution estimation on localised ‘image patches’; specialised treatment of small vessels by transformation to the one-dimensional domain to ensure enhanced detection; and excellent accuracy (93.42%) as compared with the recent active-contour based method by Al-Diri et al. [1] (92.58%) on the public DRIVE retinal image database [2].

Quantitative analysis of tendon ECM damage using MRI

There is a growing demand for non-invasive methods to diagnose tendon injuries and monitor the healing processes of their repair. One particular target is to assess the quality of tendon tissue, which requires imaging modalities, such as Magnetic Resonance Imaging (MRI), that capture structural features of the extracellular matrix (ECM). However, to date there has been limited understanding of the physiological source of intratendinous MRI signal. This paper presents a novel image analysis method, based on low level features, which capture the intrinsic structural properties in order to identify ECM damage. More specifically, continuous intrinsic dimensionality (ciD), based on local image descriptors and derived from the monogenic signal, is used to examine the disruption of 1D structures. The damage measure is summarized using confidence values derived from the bi-modality of local non-parametric probability density functions. Areas of normal and disrupted ECM are detected on MR images of healthy and damaged samples.

Sharing and Reusing Cancer Image Segmentation Algorithms Using Scientific Workflows: Pros and Cons

Image analysis researchers would benefit considerably by sharing and reusing image processing algorithms. We consider some of the issues that researchers face in trying to provide algorithms in a shareable and reusable form illustrating our approach in the context of medical imaging needs and workflow for colorectal cancer. We consider the use of workflow as a model for developing and reusing components of medical imaging and specifically we consider a solution built using .Net and Windows Workflow Foundation.

Colorectal MRI image registration using phase mutual information from non-parametric probability density function estimator

We propose a nonrigid registration algorithm and apply it to align pre- and post-chemotherapy colorectal MRI images. The algorithm combines feature-based and intensity-based image registration methods. We use local phase, as computed by monogenic signal, as the feature descriptor, and as the similarity measure in the registration algorithm, phase mutual information, which is estimated using NP windows, a non-parametric probability density function (PDF) estimator. Local deformations are modeled using the polyaffine transformation, which guarantees a smooth and invertible warping toward high image resolution. The algorithm is implemented in an adaptive manner, which makes the registration efficient and reliable. We show encouraging preliminary results and will include a performance evaluation on fives of cases in the final paper.