A Novel Paradigm for Ultrasound-based Intervention and Therapy

Probabilistic modeling of Ultrasound Radio- Frequency Time Series Data for Characterizing Cancerous Regions

Ultrasound RF time series data carry a large amount of tissue specific information that can be used in differentiating between cancerous and normal tissues. In this project, the goal is to find tissue specific signature using probabilistic modeling of RF time series data. Hidden Markov Models (HMMs) technique will be used to build a model of cancerous versus normal tissues in patients. HMMs are probabilistic models of linear sequences of observables. They are widely used in the analysis and detection of patterns in time series data. Using HMMs on RF time series can be tackled in different ways. A simplistic approach to the application of HMM is to summarize the information relayed by the RF time series data of cancerous regions of a patient in one model and the normal regions in another model. At each time point, the model shows the probability of getting an observation from different probability distributions with specific means and standard deviations calculated from a training dataset. Figure 3 shows a graphical representation of a sample HMM of RF time series data for a cancerous region.


       Figure 3 Graphical representation of a sample HMM of RF time series data

Classification Between Different Malignancy Levels of Prostate Cancer Using Ultrasound RF Time Series

Prostate cancer is one of the most common cancers diagnosed in North American men, exceeding lung cancer. The objective of this work is to evaluate the performance of RF time series on differentiating between low grade and high grade of prostate cancer. The detection of cancer is important in the diagnosis stages of prostate cancer. However, given the high life expectancy nowadays, an invasive surgery should be avoided as long as it is not necessary. Detection of the grade of the cancer or differentiation between low and high grade of cancer can help physician for this purpose. It can prevent the patient from undergoing an invasive surgery if the cancer is not high grade, which can also benefit the patient economically. Given the proven ability of RF time series in differentiating between cancerous and normal prostate tissue, here we have used this tool on characterizing multiple grades of cancer in prostate. A database of ex vivo ultrasound images of prostate tissue has been collected from 25 patients who underwent prostatectomy. Spectral based features of RF time series have been used to highlight the high grade of prostate cancer in areas as small as 1 mm x 1 mm. Gleason Score (GS) system has been used as a prognostic factor in determining the aggressiveness of prostate cancer. Cancerous areas with GS of 7 and above with primary Gleason Grade of 4 or 5 have been considered to be high grade. Using a Support Vector Machine (SVM), a classification accuracy of 81 % and an area under ROC curve of 0.88 has been achieved on a leave-one-patient out cross validation criteria. The promising results confirm the ability of RF time series on classification between multiple malignancy levels of prostate cancer.

Ultrasound-guided Characterization of Interstitial Ablated Tissue Using RF Time Series: Feasibility Study

Ablation therapy is an active field of study as a minimally invasive cancer treatment modality in the last few decades. Using this modality, the surrounding tissue can be preserved while targeting specific tumour locations. The limitation associated with ablation therapy is primarily the difficulties in monitoring the temperature rise and the extent of the ablated region in the tissue.

In this project, we present the results of a feasibility study to demonstrate the application of ultrasound RF time series imaging to accurately differentiate ablated and non-ablated tissue. We perform ex vivo ablation experiments on homogeneous chicken breast tissue specimens and in situ ablation experiments on porcine liver in the operating room immediately after an animal is sacrificed using ultrasound interstitial thermal therapy applicators (USITT). For 12 ex vivo and two in situ tissue samples, RF ultrasound signals are acquired prior to, and following, high intensity ultrasound ablation. Spatial and temporal features of these signals are used to characterize ablated and non-ablated tissue in a supervised-learning framework. In cross-validation evaluation, a subset of four features extracted from RF time series are able to produce a classification accuracy of 84.5%, and an area under ROC curve of 0.91 for ex vivo data, and an accuracy of 85% for in situ data.



Prostate Cancer Diagnosis Using Ultrasound Radio-Frequency Signals

Prostate cancer (PCa) is the most commonly diagnosed malignancy, and the second leading cancer-related cause of death in North American men. If diagnosed early, PCa can be managed with a 5-year relative survival rate above 95%. The current standard for PCa diagnosis involves ultrasound-guided core needle biopsy; however, the biopsy procedure is not scaled to individual patients due to the lack of sensitivity and specificity of conventional ultrasound images. Previously, a tissue typing approach was proposed that uses ultrasound RF time series acquired from a stationary transducer and issue position over a few seconds.


The goal is to evaluate the application of RF time series for in vivo cancer detection and for cancer treatment. I pursue this goal in the context of three experiments involving: i) prediction of cancer for in vivo radical prostatectomy cases, ii) prediction of cancer following prostate biopsy, and iii) prediction of changes to the tissue following interstitial ablation therapy. Finally, in simulation and controlled laboratory experiments, the rise of tissue temperature is explored as a potential source of tissue typing information in RF time series.


Related Publications: 
Subscribe to RSS - A Novel Paradigm for Ultrasound-based Intervention and Therapy