For ML part, please refer repository: https://github.com/diwan-kadir/CancerDetect

Project by: Kadir Diwan | Rajeev Bandi | Arfah Upade

Cancer is the second leading cause of death globally with an estimated amount of 9.6 million deaths, or one in six deaths, estimated in 2018 by the World Health Organisation (WHO) [28]. A lot of the time, prevention of diseases like Cancer, Diabetes, and Alzheimers are treated late after their onset, as detection is done late.

CancerDetect aims at giving medical professionals/researchers and patients/subjects a prediction model based on their current gene profiles that indicate the probability of the type of Cancer that they may be affected by. CancerDetect uses the patient’s gene profile to do the prediction.

Diseases such as Cancer tend to be detected after their onset which leads to infeasible prevention and difficult or painful cures. If a prediction can be given to medical professionals and their patients about the probability of the patient having cancer, the medical professional may assist the patient in taking the right steps so as to detect and prevent the first and later stages of Cancer.

CancerDetect provides a user-friendly interface to medical professionals, where they can upload the gene data of their patients after filling in the necessary details. A report is generated for the medical professional, in which the type of Cancer, the name of the Cancer, the probability percentage of the Cancer that could occur to the patient, and a remark such as “High” or “Low”, is given.

The report also presents a scatter plot for visualization purposes. The report is printable and can be given to the patient for further evaluation. CancerDetect’s probability of the occurrence of cancer within the human body works on a prediction model based on concepts of Bioinformatics, Deep Learning, and Gene Profile Reading. This prediction model enables CancerDetect to get the probability percentage that is then given to the medical professional or patient. [Refer to Section IV. PREDICTION MODEL]

For building and training Deep learning models various languages and their deep learning specific libraries are available. Cancer detect uses Python (version 3.8) and Google’s Keras Library (version compatible with python 3.8) along with other libraries such as Scikit learn, Pickle, Pandas for data processing and other miscellaneous jobs.

The frontend is hosted using Python’s Flask framework which provides flexibility as well as rapid development

Python 3.8, Keras, NumPy, Pandas, Pickle, Scikit Learn, TensorFlow

IDEs Used: Spyder, Jupyter Notebook

The TCGA dataset consists of seven thousand genes of eleven thousand individuals which in total is roughly 600 MB and training the convolutional model with 20 epochs would definitely require external graphics. The model was trained on a Graphics card having 4Gb DDR5 memory and 7 Gbps memory speed (Nvidia 1050 GPU). The prediction is not GPU-dependent as compared to training.

Abbreviation - Meaning | TCGA - The Cancer Genome Atlas | WHO - World Health Organization | AI - Artificial Intelligence | DL - Deep Learning | DNA - Deoxyribonucleic acid | CSV - Comma Separated Values | SaaS - Software as a Service | CNN - Convolutional Neural Network | IDE - Integrated Development Environment | GPU - Graphical Processing Unit

The current model predicts the cancer types according to the gene profile. Future enhancements can be:

a) Detecting the stages of cancer b) Conversion of the whole system into a SaaS module. c) Extending it to more conditions like (other diseases, Diabetes, Alzhiemers, etc.) d) Integrating the existing models with CancerDetect and providing a report which covers a large range of software-based tests and their analysis. e) Make CancerDetect available for iOS and Android.