S979 Artificial Intelligence and Digital Single-operator Cholangioscopy: Automatic Identification of Tumor Vessels in Patients with Indeterminate Biliary Stenosis

<p>The presence of plastic litters in the coastal zone has been recognized as a significant problem. It can dramatically affect flora and fauna and lead to severe economic impacts on coastal communities, tourism and fishing industries. Traditional beach litter reports include individual transects on the beach, reporting on the litter's presence through a dedicated measuring protocol. In the new era of drone imagery, a new integrated coastal marine litter observatory is proposed. This observatory is based on aerial images acquired through citizen science using low cost self-owned drones and the automatic identification of litter accumulation zones through computer vision. The methodology consists of four steps: i) a dedicated protocol for acquiring drone imagery from non-experienced citizens using commercial drones, ii) image pre-processing (image tiling and geo-enrichment) and crowdsourced annotation, iii) data classification to litter and no litter though an artificial intelligence classification approach and iv) marine litter density maps creation and reporting. The resulted density maps currently are produced calculating the tiles containing litter at areas of hundred square meters on the beach and the entire process requires some minutes to run once the aerial data is uploaded online. The density maps automatically are reported to a spatial data infrastructure, ideal for time series analysis. Classification accuracy calculated against manual identification of 77.6%. The coastal marine litter observatory presents several benefits against traditional reporting methods, i.e. improved measurement of the policies against plastic pollution, validating marine litter transportation models, monitoring the SDG Indicator 14.1.1, and most important, guiding the cleaning efforts towards areas with a significant amount of litter.</p>

Download Full-text

Development and evaluation of a deep learning based artificial intelligence for automatic identification of gold fiducial markers in an MRI-only prostate radiotherapy workflow

Physics in Medicine and Biology ◽

10.1088/1361-6560/abb0f9 ◽

2020 ◽

Vol 65 (22) ◽

pp. 225011

Author(s):

Christian Jamtheim Gustafsson ◽

Johan Swärd ◽

Stefan Ingi Adalbjörnsson ◽

Andreas Jakobsson ◽

Lars E Olsson

Keyword(s):

Artificial Intelligence ◽

Deep Learning ◽

Automatic Identification ◽

Fiducial Markers ◽

Prostate Radiotherapy

Download Full-text

How much deep learning is enough for automatic identification to be reliable?

The Angle Orthodontist ◽

10.2319/021920-116.1 ◽

2020 ◽

Vol 90 (6) ◽

pp. 823-830

Author(s):

Jun-Ho Moon ◽

Hye-Won Hwang ◽

Youngsung Yu ◽

Min-Gyu Kim ◽

Richard E. Donatelli ◽

...

Keyword(s):

Artificial Intelligence ◽

Deep Learning ◽

Random Sampling ◽

Sampling Method ◽

Automatic Identification ◽

Logarithmic Scale ◽

Data Sets ◽

Radial Error ◽

Optimal Quantity ◽

Learning Data

ABSTRACT Objectives To determine the optimal quantity of learning data needed to develop artificial intelligence (AI) that can automatically identify cephalometric landmarks. Materials and Methods A total of 2400 cephalograms were collected, and 80 landmarks were manually identified by a human examiner. Of these, 2200 images were chosen as the learning data to train AI. The remaining 200 images were used as the test data. A total of 24 combinations of the quantity of learning data (50, 100, 200, 400, 800, 1600, and 2000) were selected by the random sampling method without replacement, and the number of detecting targets per image (19, 40, and 80) were used in the AI training procedures. The training procedures were repeated four times. A total of 96 different AIs were produced. The accuracy of each AI was evaluated in terms of radial error. Results The accuracy of AI increased linearly with the increasing number of learning data sets on a logarithmic scale. It decreased with increasing numbers of detection targets. To estimate the optimal quantity of learning data, a prediction model was built. At least 2300 sets of learning data appeared to be necessary to develop AI as accurate as human examiners. Conclusions A considerably large quantity of learning data was necessary to develop accurate AI. The present study might provide a basis to determine how much learning data would be necessary in developing AI.

Download Full-text