In our own words briefly describe what is the ‘Cryo-EM’ technology for imaging? What type of images is generated by help of this technology?

  Cryogenic electron microscopy (Cryo-EM) is one of the techniques, which is related to the technique of electron microscopy (EM), and its application is made on those samples, which are cold as per cryogenic temperatures, and they are used in vitreous water’s environment. The application of a solution of the aqueous sample is made on the plunge-frozen and grid-mesh with liquid ethane. An electrons beam is used by the transmission electron microscopes (TEMs) so that materials and molecules structure is examined when they are located at the atomic scale. When a thin sample observes going through a beam, the interaction of the beam is made with the molecules, and then the sample’s image is projected on the screen of the projector (CCD). The electrons' wavelength is quite shorter as compared to the wavelength of the light; that’s why the detail revealed by the electrons is much finer as compared to the details shown by super-resolution light microscopy (BROADWITH, 2017)

b) What are some of the main benefits of the ‘Cryo-EM’ technology for the medical and biological fields?

      The ‘Cryo-EM’ technology comes with so many benefits for the medical and biological fields. The first great benefit of ‘Cryo-EM’ technology is that the required samples are not large, rather very small when its structure is to be determined. Other imaging technologies require large sample sizes. Moreover, the hydrated state of the sample is preserved because actual fixation and rapid freezing are performed in vitreous ice. So, the state of any sample can be seen without any issue. Moreover, the samples in this technique can be viewed in a wide range, which makes things easier for the medical processes. In addition to that samples are inhomogeneous, which means that they come with a high level of magnification, which helps to closely study the specimen. With the help of ‘Cryo-EM’ technology, the chemical environment can easily be controlled which allows examining molecules more effectively (microscopemaster.com, 2019)

c) What is EMPIAR? What are the smallest and the largest sizes of datasets available at this archive?

        The term EMPIAR is derived from the words “Electron Microscopy Public Image Archive”. It is actually a free public resource, and it is used for raw images as well as images based on 2D electron microscopy. The great thing about this archive is that both the smallest and largest datasets are available here, which can be used for various purposes. For instance, a user can download, upload or browse raw images, which can be helpful in building a 3D structure. It is important to mention here that 2D images and datasets provided by EMPIAR are very critical to develop the structure for molecular machines as well as for biomacromolecules. The Electron Microscopy Data Bank (EMDB) is complemented by EMPIAR. The largest datasets in this archive can go up to the level of terabyte size, which shows its depth for an archive (EMBL-EBI, 2019)

d) References for Question 1: You must properly cite at least 2 references for your answer.

References for Q.1

BROADWITH, P. (2017). Explainer: What is cryo-electron microscopy. Retrieved November 28, 2019, from https://www.chemistryworld.com/news/explainer-what-is-cryo-electron-microscopy/3008091.article

EMBL-EBI. (2019). What is EMPIAR? Retrieved November 28, 2019, from https://www.ebi.ac.uk/training/online/course/empiar-quick-tour/what-empiar

microscopemaster.com. (2019). Cryo-Electron Microscopy. Retrieved November 28, 2019, from https://www.microscopemaster.com/cryo-electron-microscopy.html

Question – 2

a) Name the Pattern Recognition tasks and the Anatomical Regions that the authors have considered for Medical Image Analysis in their work

            It is important to understand that authors have taken a great approach for analyzing medical imaging with deep learning. They used different pattern recognition tasks in relation to the Anatomical Regions. It is vital to look at them one by one. In total, four pattern recognition tasks were used, and the first one was detection/localization. The second major pattern recognition task was segmentation. The third and fourth pattern recognition tasks in this paper were registration and classification. Keeping these pattern recognition tasks in view, the authors used six human anatomical regions. The first three important human anatomical regions were brain, breast, and eye. The other three human anatomical regions were chest, abdomen, as well as miscellaneous. So, this categorization was used in this research article.

b) Briefly describe in your own words the DL Method of Localization/Detection discussed by the authors as applied to Eye Disease diagnosis.

            It is evident in the research article that authors have used different detection methods for the diagnosis of different diseases related to different human anatomical regions such as the eye. When the eye section is analyzed closely, it shows that authors have discussed deep learning (DL). They discussed a DL model, which has its basis on the inception architecture so that Glaucomatous Optic Neuropathy (GON) can be identified with regards to retinal images. The AUC achieved by this particular DL model was 0.986, which proved helpful in the distinction of GON eyes from healthy eyes. For diagnosing eye diseases, the authors also discussed the topic of deep transfer learning method, which also proved beneficial.

c) Briefly describe in your own words the DL Method of Image Segmentation discussed by the authors as applied to Brain Disease diagnosis.

            When the segmentation section was discussed in detail in this research article, the authors came up with different methods of the diagnosis to talk about. So, they also used deep learning (DL) method and discussed its relevant aspects. They talked about the DL technique, which can be used for the segmentation of the brain tumor. For this purpose, proper integration of Conditional Random Fields (CRFs) and Fully Convolutional Networks (FCNs) is needed to combine a framework, which is helpful in achieving segmentation. It is important to mention here that three trained segmentation models were used with the help of 2D image slices as well as patches. They also performed training both for FCN as well as CRF.

d) Describe in your own words at least 2 challenges faced by DL as applied to Medical Image Analysis that the authors have discussed in their work

            It is vital to understand that when deep learning (DL) is applied for the analysis of medical imaging, it not only provides opportunities, but it also comes with some challenges as well. The authors have talked about the different challenges faced by DL. The first major challenge in their view is that data for deep learning in the medical field is not properly annotated. The proper annotation is essential so that a powerful deep model can be learned with suitable training data; otherwise, things will remain complex. The second major challenge faced by deep learning as per the authors is that data is extremely imbalanced. The samples being used in the datasets of medical imaging are imbalanced. The authors used an example that when datasets for identification of breast cancer are used, the majority of the samples are negative, which means that samples don’t come with a considerable balance to get deep learning on the issue (ALTAF, ISLAM, AKHTAR, & JANJUA, 2019)