Self-driving cars & lung cancer diagnosis quandary

From a vantage standpoint, one could hardly surmise the uncanny resemblance in between self-driving cars & the diagnosis of lung cancer. Only when one starts to dive deeper into the research going in both the fields, one is profoundly enlightened by the challenges & the opportunities.

Cars were, are & will remain the bespoke suits of the transportation industry. One can exercise the luxury, comfort & elegance as one may desire. No other means of transport has been able to replace it, irrespective of our strive for environmental conscientious. Self-driving cars are the proclaimed future of the wheel-based transportation with flexibility of the personalised transport & yet supporting the idea of mechanized interaction of multiple vehicles. It culminates the dream of smooth flowing traffic into a foreseeable reality.

Contrarily, lung cancer diagnosis mundane as it may sound has a discernible relevancy to people lives. Based on the number of cigarettes consumed per capita (As per WHO: Over 15 billion per day) & increasing longevity of human, one can only suggest the potential possibility of diagnosis after a certain age. It excludes the people exposed to smoke & gas based on occupational hazards & transportation. Initial stage state of the art diagnosis of lung cancer relies heavily on analysing CT scan images & identifying the potential malignant tumour. That too only when clinical symptoms start to show up. Thus, most of the successful diagnosis just happen much too late, limiting the potential survival chances. Lung cancer is also the form of cancer with the highest mortality rate in the world.

The future though of both fields is converging into a realm that requires solving some of the fundamental challenges faced by some of the smartest engineers today. Self-driving cars relies heavily on gathering huge chunk of data from many different types of sensors. The processing of the huge data at real-time without error has become the true cradle of the problem here. Multi-layered processing of the data at real-time is processing expensive & thus also manifold increasing the complexity of the software.  Increased complexity does not help the cause of human safety. Engineers have started looking towards artificial intelligence (AI) as a potential field to find patterns from the past runs & approximating them to make advanced predictions. Although lung cancer diagnosis does not demand real time processing of the huge amount of the data, but a successful advanced prognosis of the disease has pushed some biomedical engineers towards AI. It involves learning the CT scans of many advanced lung-cancer patients & correlating them for an advanced diagnosis with potential healthy ones.

AI is a fledgeling field & new algorithms are arising now & then to approximate the past data patterns as accurately as possible. There are already some non-critical successful applications in place affecting each of us in our daily life which includes image processing in our cell phone cameras. Even the tools used to implement, optimize the data are relatively new technologies. One can honestly estimate the limitations of AI when one tries to communicate with their virtual assistants on phones or home automation devices. The inability of virtual assistant to grasp human emotion behind the said expression limits the ability to create a 2-way communication.

Though the fundamental problem that it must solve is not about how much data can be processed in real-time but on how accurate it can predict & take predictive actions from the past learnings. In both self-driving cars & the lung cancer diagnosis the focus is to reduce human error & to improve upon the overall safety of the passenger or patient. Even though a human brain might be the best-known computer; still the memory it can retain & the sensors human have been much more limited in amount excluding the sense of touch.

Moving forward with an AI that thinks & behaves like human has its own pitfalls. Humans make error. Be it analysing CT scans for lung cancer diagnosis (unwarranted diagnosis & surgeries) or driving a car (varying concentration). Both could be fatal. On one hand these errors can be minimized by giving up control & handing it over to a computer with a focused task. On the other, AI is unable to mimic the capability of a human being and react to an unpredictable situation with human’s best interest. The future is interesting with a huge potential for AI but the cost of such an exercise also needs to be prepensed. Thus coinciding with the words of wise men, ‘Every cloud has a silver lining’.

References

1. https://www.who.int/tobacco/en/atlas8.pdf?ua=1 Cigarette Consumption
2. https://tobaccoatlas.org/topic/consumption/ Tobacco consumption
3. https://gis.cdc.gov/Cancer/USCS/#/AtAGlance/ US CDC: Cancer Statistics At a Glance
4. https://link.springer.com/article/10.1007/s11042-019-08394-3 Deep learning for lung cancer detection and classification
5. https://pubmed.ncbi.nlm.nih.gov/22298307/ Identification of candidate biomarkers for early detection of human lung squamous cell cancer by quantitative proteomics
6. https://pubmed.ncbi.nlm.nih.gov/31754135/ Deep segmentation networks predict survival of non-small cell lung cancer
7. https://xgboost.readthedocs.io/en/latest/ XGBoost Documentation
8. https://lightgbm.readthedocs.io/en/latest/index.html Welcome to LightGBM’s documentation!
9. https://pubmed.ncbi.nlm.nih.gov/33803033/ Predicting Proteolysis in Complex Proteomes Using Deep Learning
10. https://ieeexplore.ieee.org/document/9244647 Voronoi-Based Multi-Robot Autonomous Exploration in Unknown Environments via Deep Reinforcement Learning
11. https://arztsamuel.github.io/en/projects/unity/deepCars/deepCars.html Deep Learning Cars
12. https://www.youtube.com/watch?v=UkYxKu2pllw I programmed an A.I. for Need For Speed: Most Wanted
13. https://www.youtube.com/watch?v=1W9q5SjaJTc Ride in NVIDIA's Self-Driving Car
14. https://www.eea.europa.eu/data-and-maps/data/external/annual-distance-travelled-by-cars Annual distance travelled by cars
15. https://towardsdatascience.com/applied-deep-learning-part-1-artificial-neural-networks-d7834f67a4f6 Applied Deep Learning - Part 1: Artificial Neural Networks