Diagnostic Imaging
Medical Diagnostic Imaging and Machine Learning
Imaging and machine learning is a new emerging field which may help train doctors and diagnose patients. Current tryouts are quite promising, but further investments are needed to get into the high quality results that human medicine requires. Medical image diagnostics have been tested to diagnose pneumonia and breast cancer, as examples. In pneumonia there are two types of pneumonia, one caused by virus and another by bacteria. Treatments are quite different one from the other so it is important to differentiate them. This technology could help as it is hard to see any difference for the non trained eye.
Statistical Results ML for Pneumonia
ML Results to detect and differentiate virus pneumonia and bacteria pneumonia were interesting. The initial model was trained with about 2,700 images and shows its accuracy may improve when more images are used. The dataset was a public dataset.
Breast Cancer Imaging and Machine Learning
Breast Cancer detection with machine learning is a new emerging field which may also be useful for doctors and hospitals. Cancer is shown as opaque, a more dense substance within the image taken.
Statistical Results ML Breast Cancer
About 8,000 augmented images were used of a public dataset to train a machine learning model to distinguish breast images with cancer to the ones without cancer, the results were quite impressive. It would be interesting to run the algorithm again with other sets to verify that ML algorithms can indeed identify quite well images which contain cancer from the images which show healthy breasts.
Thermal Scanning
Thermal scanning feeding images to a machine learning algorithm can help find out if people are in the area or if there's a higher temperature than expected in a process (fire, malfunction). Imaging may be used to set up alarms
Flotation Froth Control in Mineral Processing
Froth in flotation cells is hard to control and hard to monitor. Many times, workers need to be monitoring if processes are working. Machine learning algorithms have proven in studies that they can recognize if there's froth or not, keep conditions to keep the froth going and set an alarm if intervention is needed as part of a full process control system.
Thermal Imaging in Glass Annealing
Thermal imaging in glass annealing is a critical, non-contact process control method using infrared cameras (often at 5micro m wavelengths) to monitor surface temperature uniformity during slow controlled cooling.If hot/cold spots are detected that prevent the glass from been between 900-960F, alarms may sound or it may modify the process to remove such spots. Cracks may be spotted.
Wafer Inspection
It is quite important to be monitoring the wafer lithography process as wafers are expensive. This process can be automated a lot by using machine learning as defects typically leave lower layers uncoated or overcoated. Each one of the layers of the wafer has different colors, so a system that detects difference in colors in certain parts of the wafer can fully detect errors. If the drawing of the wafer carvings are used to create a mask for the picture, we can make sure that we remove all the color in the areas that had to be carved out and vice versa for the other area (negative mask) . By reviewing both sets, we can determine if an area was not carved out or if an area that shouldn't be coated still is coated. That can be done for each step of the coating/carving process.