Abstracts Submitted: 510
Number of Users: 710
View Abstracts Submitted
Back to home Page
Prostate cancer is a category of cancer which occurs in men due to uncontrolled multiply of prostate gland tissue. The resultant increase in size of prostate gland may disturb the normal flow of urine because of the squeezing of urethra, a urine carrier, passing through the centre of prostate. The usual screening methods for prostate cancer includes prostate specific antigen level check in blood and a digital rectal examination for defining the morphology of prostate gland. The suspicious results indicated by these two results may lead to a Transrectal ultrasound (TRUS) guided biopsy as a measure of confirmation of cancer. Although TRUS guided biopsy is the gold standard for prostate cancer detection, due to the mechanical property based contrast of ultrasound imaging, instead of directed sampling of prostate gland, the whole gland has to be scanned systematically for tissue collection. This process increases patient discomfort and sometimes it misses the malignant region of the prostate gland. A new imaging modality, capable of providing functional information of tissue, is required for better detection of prostate cancer. Photoacoustic imaging (PA) is an emerging hybrid medical imaging modality which exploits optical property based contrast of pure optical imaging to provides functional information of soft tissue. Unlike pure optical imaging, it does not suffer from low resolution inside the tissue specimen. Machine learning is a field of computer science in which the computer systems are trained on the statistics of given data to impower these systems to make inference on the new incoming data. In medical field, machine learning algorithms like Artificial neural network, support vector machine and random forest classifier allowed the development of computer aided diagnosis (CAD) systems which serves as a second opinion to the physicians. In this work we have attempted a computer assisted prostate cancer detection method for differentiating malignant, normal and benign prostatic hyperplasia (BPH) prostate tissue using photoacoustic imaging and machine learning. The photoacoustic imaging is done on the freshly excised human prostate specimen to collect the PA signals for different operating wavelengths of laser. The PA signals corresponding to three region of interests i.e malignant prostate tissue, normal prostate tissue and BPH as confirmed by pathologists were used to form the mutually exclusive training and testing set. A random forest classifier is trained on the training set and is then evaluated for classification performance on testing set. We consider binary classification problem on two cases, one being classifying the malignant prostate tissue from normal prostate tissue and other is classifying malignant prostate tissue from non- cancerous prostate tissue where the later contains the combination of BPH and normal prostate tissue. The classification results in terms of sensitivity and specificity also the receiver operating characteristic (ROC) curve analysis were recorded. For malignant versus non-cancerous tissue classification on 760nm wavelength data we achieved average sensitivity and specificity as 0.89 and 0.91 respectively, while for malignant versus normal tissue classification these numbers read as 0.94 and 0.87 respectively. These results shows that using random forest algorithm with raw time dependent PA samples it is possible to detect prostate cancer with high sensitivity.
© Copyright 2017 All Rights Reserved