Embedded Files
Quantitative Photoacoustic Tomography Using Deep Neural Networks
Quantitative Photoacoustic Tomography Using Deep Neural Networks
Photoacoustic tomography (PAT) is an imaging modality that benefits from high spatial resolution using ultrasound waves and high contrast by optical waves. This imaging modality is non-invasive and its usage is growing. In PAT, light (mostly visible or near-infrared) is emitted toward a tissue and depending on tissue's optical properties, the light is absorbed in a specific way. When the light is absorbed and the entered energy is distributed, an initial pressure arises those results in propagating ultrasound waves. The waves are received by transducers. So, the light makes an initial pressure and the initial pressure makes ultrasound waves. Now there are two inverse problems: one it to reconstruct the initial pressure and the other to obtain optical properties of the tissue based on how the light has been absorbed. The latter is called Quantitative Photoacoustic Tomography (QPAT). QPAT has different applications in medicine including determination of cancerous tissue, blood oxygen saturation and studies about functionality of the eye and the brain. The following is a photoacoustic image of a mouse brain:
This inverse problem has two main points. The first is that it is non-linear and the second is that preparing labeled data is hard, expensive and in some cases not possible. The first point makes deep neural networks as suggested solutions and the second means that unsupervised or semi-supervised methods may be good choices for the inverse problem. We have designed a simulation environment of a tissue and two blood vessels with different values and parameters including different levels of oxygen, different vessel sizes and different skin properties to make a versatile dataset. For each set of properties, light at different frequencies will be emitted toward the tissue to make different mixed results.
We have proposed a neural network that mimics the physics of the problem and is unsupervised which has to do the unmixing procedure. The proposed method is under assessment and may become semi-supervised in future. Anyway, the base of the proposed network structure is as the following image:
General Steps of Quantitative Photoacoustic Tomography
Diagnosis of bone fractures using a causal neural network
Diagnosis of bone fractures using a causal neural network
Bone fractures are one of the most important medical issues that require experienced specialists to diagnose, so deep neural networks can be of great help in both diagnosis and treatment in this area, also causality learning methods can increase the accuracy and speed of neural networks.
This method helps medical centers that do not have experienced specialists to diagnose faster and more accurately, and also can provide specialists with suggestions for better treatment.
Classification of Cervical Cells Using Deep Learning
Classification of Cervical Cells Using Deep Learning
Cervical cancer is a common disease among women. Prompt and timely diagnosis can help treat and prevent its progression, so screening for cervical cancer is very important. Classifying cells in Pap smear images is very challenging, and current cytological procedures have limitations for early and accurate screening for cervical cancer. Therefore, our goal is to design a deep network to classify and screen cervical cells to diagnose and predict cervical cancer.
We use morphology to detect cervical cells.
Morphology is a set of image processing operations that processes images based on shapes. The morphological operation applies a structural element to an input image and creates an output image of the same size.
Page updated
Report abuse