↓ Skip to main content

Papers

Table of Contents

Here you can find the papers for some of the projects that I have done.

VisiCell
#

VisiCell is an web and mobile application that automates counting of cells dyed with trypan blue. For its development, I have created a novel dataset consisting of 792 images of trypan blue-stained MCF7 breast cancer cells, labelled as dead or alive with bounding boxes. Trained Scaled YOLOv4 models on this dataset, achieving an accuracy of 0.958 in the mAP.5 metric. Created a user-friendly web and cross-platform mobile application using Flask and React Native for accessibility and usability.

Glacier Mass Change Prediction (Polar Research Projects Competition)
#

3rd place in Turkey

Trained ten regression models to predict glacier mass change in Antarctica over the next decade using five independent variables. Employed the Prophet procedure to forecast independent variables for the next ten years, predicting glacier mass change with the developed regression models. Achieved a Root Mean Square Error of 45.98 using the Ridge regression model.

Disease Detection From Abdominal CT Scan (Teknofest)
#

🥇1st place in Turkey

Developed a machine learning model network comprising of six YOLOv4-P6 models for detection of appendicitis, cholecystitis, pancreatitis, kidney and bladder stone, diverticulitis, aortic aneurysm and dissection from abdominal CT scans. Achieved 0.743 accuracy on the mAP.5 metric on a data set provided by the Ministry of Health. Achieved first place in a national competition among high school teams in Turkey.

Breast Cancer Detection From Mammograms (Teknofest)
#

3rd place in Turkey

Trained two EfficientNet-B3 models to predict BI-RADS categories and breast composition from mammograms. Trained model achieved an accuracy of 0.6713 on a given test dataset of 3,972 patients.

Labtoolkit
#

Developed an original classical image processing pipeline for detecting live and dead cells, utilizing techniques such as Non-local Means denoising, guided image filtering, CLAHE, Otsu thresholding, morphological operations, distance transform, watershed segmentation, and contour counting. Achieved 71% accuracy in live cell detection and 53% accuracy in dead cell detection.