Dear Editor,
With great interest, we read the article in press, published by Sanyal et al1 in this journal. The authors have done a commendable work by developing a high-performance convolutional neural network (CNN)–based tool for detecting cervical cancer. CNN is an artificial intelligence (AI) algorithm which enables a machine to accurately classify smears as ‘normal’ or ‘abnormal’. Diagnostic accuracy of 95.46% over a validation set of 441 images as reported in the study is appreciable. The performance of this AI tool needs to be evaluated over many smear samples collected from various hospitals spread across the country before using it as a screening tool. We wish to signify how the trending technology – cloud computing – shall prove beneficial in evaluation and further usage of this tool.
Explained in simple words, cloud computing offers an easy way to store and share data via internet (such as e-mail). Cloud facilities such as Amazon Web Services and Microsoft Azure are being offered by big software companies such as Amazon and Microsoft, respectively, for improving the healthcare sector.2 Authors are suggested to make use of any of these services to enable easy access to the CNN model via internet. The authors have used Python programming language in their research work and are suggested to use ‘Flask’ and ‘Pickle’ software libraries for deployment.
Once hosted on cloud, the CNN model can be accessed via internet anywhere across the globe by authorized hospitals and diagnostic centres. Cervical smear pools collected by these healthcare centres should be analysed by pathologists working there and by the CNN model (via internet). Match/mismatch between the findings of pathologists and the CNN shall help in easy evaluation of the CNN tool. Modifications in the algorithm can be performed to optimize the performance.
Once the performance of the CNN tool is found to be satisfactorily good, it can be made available (via internet) to healthcare centres in areas facing shortage of medical facilities, skilled pathologists and doctors. In such places, the disease usually remains undetected and leads to mortality. Images of a cervical smear can be easily uploaded in the CNN tool by the authorized medical practitioner working in such healthcare centres. In the absence of skilled pathologists, images shall be analysed by the CNN tool to detect abnormality in the smear. The patient need not wait for many days for the results as AI tools take very less time in processing the images. The patient diagnosed with abnormality in the smear shall be recommended to seek medical help for further investigation and treatment.
Similar AI tools accessible via internet are already being used by primary healthcare centres in the USA for diagnosing diabetic retinopathy.3 However, India lacks in this regard. AI tools are not a substitute for doctors/pathologists. These tools should be considered as a technological aid for doctors as these can help significantly in early diagnosis of the disease. AI-based healthcare tools available via internet shall make healthcare facilities quite accessible and affordable in rural parts of the country.
Through this letter, we wish to again congratulate the authors for their brilliant research. We also appeal to premier medical colleges and hospitals to sync with engineering colleges and software companies to promote similar research for various critical diseases such as cardiovascular diseases, tuberculosis, and so on.
References
- 1.Sanyal P., Ganguli P., Barui S. Performance characteristics of an artificial intelligence based on convolutional neural network for screening conventional Papanicolaou-stained cervical smears. Med J Armed Forces India. 2019 doi: 10.1016/j.mjafi.2019.08.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.MSV J. 2020. The Rise of Artificial Intelligence as a Service in the Public Cloud [Internet]. Forbes.https://www.forbes.com/sites/janakirammsv/2018/02/22/the-rise-of-artificial-intelligence-as-a-service-in-the-public-cloud/#6dddfa5e198e [cited 6 September 2020]. Available from: [Google Scholar]
- 3.IDx-DR . Docwire News; 2020. The First FDA-Approved AI System, Is Growing Rapidly - Docwire News [Internet]https://www.docwirenews.com/docwire-pick/future-of-medicine-picks/idx-dr-the-first-fda-approved-ai-system-is-growing-rapidly/ [cited 31 January 2020]. Available from: [Google Scholar]
References
Reply to ‘Artificial intelligence and improved healthcare’.
Dear Editors,
Thank you for your observations. Our work has progressed to where we evaluate the Performance of a multi class classifier based on a convolutional neural network in recognition of normal tissue from histological sections and will be publishing the same, maybe in this esteemed journal. In this experiment a total of 677 histologic images were micro photographed, split into two subsets, training (205) and validation (472 images). and after validation, we performed a pilot run of the model on 23 images with known pathology. Though exciting, we realize that it is best suited in a repetitive kind of work where fatigue can overcome the sharp eye of an average pathologist. We have also realized that the pathologist can't be replaced and needs to be at the center of this process. Interestingly we have used the Fast-AI platform. With due permissions and concerns in place we would also like to collaborate in the future with like minded teams. I thank you for your kind words which will motivate us further.