Introduction
The top areas of cancer computer-aided diagnosis (CAD), which seem to be most often discussed and particularly well translated to practice, are breast, lung, and colon cancer diagnosing (Giger, 2015, p. 409). With respect to machine learning (ML), Giger (2015) suggests that it is applicable to the case of CAD since CAD stems from computers being capable of learning directly from image data (p. 413). Similarly, data mining (DM) is also applicable to CAD: Giger (2015) states that CADs can use it to retrieve images, and Gillies, Kinahan, and Hricak (2016) demonstrate that DM is required to transform CAD data into diagnostic models (p. 565).
Giger (2015) shows that ML helps to improve the quality of CAD since the ability of a computer to study and learn from multiple images enables specialists to “train” CADs in performing tasks like detecting malignant nodes in human lungs (p. 411). Moreover, the approach is also developing: Cheng et al. (2016) discuss deep learning, which is an extended version of ML that is characterized by easier and more direct training data analysis (p. 2). Thus, the key elements of CAD are closely connected to ML and DM, which are used to encourage CAD’s practical application while also promoting research in CAD and ML.
Concerning clinical outcomes, CAD has been shown to improve the sensitivity of radiologists while also providing good results with ultralow radiation doses (Messerli et al., 2016). Messerli et al. (2016) point out that early diagnosing is crucial for improving clinical outcomes for cancer and mention that the reduction of radiation dose by 92% is a positive outcome, which makes these findings significant for practice.
What are the top areas of use, with regard to cancer, for CAD?
Complete automation would make CAD more convenient and improve the speed of diagnosing. This development is especially attractive because of the problem of understaffing, which, in my experience, is quite noticeable in US healthcare. Therefore, the lack of automation does seem to be a restriction that impacts CAD usage. However, the lack of CAD automation can be explained by the fact that CAD is used to double-check diagnoses, which should result in more accurate diagnoses (Cheng et al., 2016) and would be expected to make the exclusion of humans from diagnosing rather undesirable. Moreover, CAD is not error-free (Hambrock, Vos, Kaa, Barentsz, & Huisman, 2013, p. 526). Nowadays, full CAD automation does not appear to be reasonable.
How are machine learning and data mining algorithms leveraged with regard to CAD? What are the key steps in this process? How will these techniques, more specifically, advance research?
Several healthcare fields seem to benefit from currently existing CADs. As shown by Hambrock et al. (2013) and Acharya et al. (2015), magnetic resonance imaging and electroencephalogram signals can be misleading, which means that the application of CAD to them is beneficial. Also, Hambrock et al. (2013) show that CAD improves the performance of specialists who are not very experienced while also being useful for experienced ones. As for specific diagnoses, Giger (2015) suggests that the diagnostics of lung and colon diseases and breast cancer is particularly well-developed in practice and research (p. 409). Thus, it can be implied that existing CADs can be of great use in various areas of healthcare.
What are some clinical outcomes that have been improved with regard to the use of CAD in clinical practice?
Eric’s presentation is well-organized, and it appears to include all the most important aspects of genomics. Given the fact that genomics is a very promising field (Cui, 2015), it is engaging to listen to the information about its history as well as its present. Despite paying attention to the notable contributors of the past, Eric manages to also introduce a discussion of modern-day genomics from three perspectives: that of benefits, challenges, and clinical outcomes. The overview is concise, however, which is why it is possible to expand it. I would appreciate a more extensive practice-oriented discussion, for example, on the current state of the translation of clinical-genomic data integration in practice. Alternatively, discussing a couple of examples of the improvements in clinical outcomes could be very engaging.
References
Acharya, U. R., Sudarshan, V. K., Adeli, H., Santhosh, J., Koh, J. E., & Adeli, A. (2015). Computer-aided diagnosis of depression using EEG signals. European Neurology,73(5-6), 329-336. Web.
Cheng, J., Ni, D., Chou, Y., Qin, J., Tiu, C., Chang, Y.,… Chen, C. (2016). Computer-aided diagnosis with deep learning architecture: Applications to breast lesions in US images and pulmonary nodules in CT scans. Scientific Reports,6, 24454-1-24454-13. Web.
Cui, J. (2015). Genomic data analysis for personalized medicine. In C. K. Reddy & C. C. Aggarwal (Eds.), Healthcare data analytics (pp. 187-218). New York, NY: CRC Press.
Giger, M. L. (2015). Future perspectives. In Q. Li & R. Nishikawa (Eds.), Computer-aided detection and diagnosis in medical imaging (pp. 409-415). New York, NY: Taylor & Francis.
Gillies, R., Kinahan, P., & Hricak, H. (2016). Radiomics: Images are more than pictures, they are data. Radiology, 278(2), 563-577. Web.
Hambrock, T., Vos, P. C., Kaa, C. A., Barentsz, J. O., & Huisman, H. J. (2013). Prostate cancer: Computer-aided diagnosis with multiparametric 3-T MR imaging—Effect on observer performance. Radiology,266(2), 521-530. Web.
Messerli, M., Kluckert, T., Knitel, M., Rengier, F., Warschkow, R., & Alkadhi, H,… Bauer, R.W. (2016). Computer-aided detection (CAD) of solid pulmonary nodules in chest x-ray equivalent ultralow dose chest CT – first in-vivo results at dose levels of 0.13mSv. European Journal Of Radiology, 85(12), 2217-2224. Web.