Arafat, Yeasin and Salam, Md Abdus and Akter, Farhana and Hossain, Mohammad Nazmul and Urmi, Amena Khatun and Afrin, Sadia and Rowshon, Mantaka and Rahman, Md Moshiour and Rahman, Nur (2026) Artificial Intelligence-Driven Medical Imaging and Predictive Analytics for Early Tumor Detection, Classification, Treatment Response Assessment, and Disease Progression Monitoring. Power System Technology, 50 (4). pp. 974-990. ISSN 1000-3673

Text Artificial+Intelligence-Driven+Medical+Imaging+-+Current+Issue.pdf Download (493kB)

Abstract

Cancer outcomes depend heavily on how early a tumor is found, how accurately it is characterized, and how quickly change sin its behavior are recognized during therapy. Medical imaging is the principal source of this information, yet the volume and complexity of imaging data now exceed what human readers can interpret consistently. Artificial intelligence (AI), especially deep learning and radiomics, offers a way to extract quantitative, reproducible information from images and to convert it into predictions that support clinical decisions. This article reviews the evidence for AI across four stages of the oncology pathway:earlydetection, classification, treatment response assessment, and disease progression monitoring. It then develops a compact mathematical framework for each stage, covering convolutional detection models, receiver operating characteristic (ROC) analysis, Bayesian interpretation of screening results, tumor growth and treatment response kinetics, and survival modeling with a Cox proportional hazards formulation. Worked calculations show that (a) a classifier with 94% sensitivity and 92% specificity has a positive predictive value of only 10.6% at 1% prevalence, (b) detecting a lesion at 0.5 cm rather than 1.0 cm gives three tumor doublings of lead time, (c) a 30% decrease in diameter underRECIST 1.1 corresponds to a 65.7% decrease in volume, and (d) a high-risk imaging signature with a hazard ratio of 2.5 shortens median progression-free survival from 34.7 to 13.9 months in the illustrative model. Five graphs and three tables summarize the results. The model outputs are illustrative simulations built from stated assumptions and are not patient data. The article closes with a discussion of validation, bias, explainability, and implementation barriers.

Item Type:	Article
Uncontrolled Keywords:	artificial intelligence; deep learning; radiomics; medical imaging; early cancer detection; treatment response; RECIST; predictive analytics; survival modeling
Subjects:	A General Works > AI Indexes (General)
Depositing User:	admin eprints
Date Deposited:	08 Nov 2026 13:50
Last Modified:	08 Nov 2026 13:50
URI:	http://eprints.umsida.ac.id/id/eprint/17509

Actions (login required)

View Item