Research story · published
Artificial Intelligence in Early Cancer Detection: A Paradigm Shift in Oncology Diagnostic
The authors report promising CNN performance, but missing methods, internal inconsistencies, and citation problems prevent confident clinical interpretation.
01
Why this matters
Pakistan faces shortages of specialist imaging staff, fragmented digital infrastructure, and delayed cancer diagnosis. A credible multicentre evaluation of AI-assisted imaging could be highly relevant. Faster triage might help specialists prioritize suspicious cases. The importance of the question, however, does not establish the reliability of the reported answer.
02
Study snapshot
- Design
- Reported mixed-methods retrospective imaging analysis plus staff interviews
- Sample
- 1,200
- Population
- More than 1,200 reported images and interviews with 13 radiology, oncology, and IT staff
- Setting
- Three named Pakistani hospitals; reported data period 2018 to 2023
- Measures
- Mammography, CT, MRI, histopathology, CNN, decision tree, and semi-structured interviews
- Analysis
- Reported 70% training, 15% validation, and 15% testing split; essential model and patient-level details are absent
CNN performance reported by the authors. No confidence interval, test-set count, confusion matrix, threshold, or external validation is provided.
03
Results, without the jargon
The authors report that their CNN classified images more accurately and quickly than a decision tree. Those values sound promising, but the paper does not provide enough detail to determine whether the model learned genuine cancer patterns, memorized site or patient characteristics, or would work on new patients elsewhere. This is preliminary reported evidence and should not be described as proof that AI improves cancer detection in Pakistan.
Reported CNN sensitivity
No confidence interval, case count, threshold, or confusion matrix is shown.
Source: PDF Table 1, printed pages 4490 to 4492Reported CNN specificity
Reported without uncertainty or a reproducible test-set description.
Source: PDF Table 1, printed pages 4490 to 4492Reported CNN accuracy
The unit of analysis and patient-level split are unclear.
Source: PDF Table 1, printed pages 4490 to 4492Reported decision-tree AUC
CNN comparison is called significant, but no statistical comparison is reported.
Source: PDF Table 1, printed pages 4491 to 4492Highest reported site accuracy
Aga Khan University Hospital; site sample sizes and external validation are absent.
Source: PDF Table 2, printed page 4492Interview participants
6 radiologists, 4 oncologists, and 3 IT administrators; qualitative reporting is limited.
Source: PDF printed pages 4490 to 449104
What this could mean
If independently validated, AI could help prioritize suspicious images and reduce turnaround time. Deployment would still require human oversight, training, secure data systems, clear accountability, bias monitoring, external validation, and prospective evaluation. Diagnostic accuracy alone does not establish improved survival, cost effectiveness, fairness, or patient safety.
05
Read this with the limitations
This retrospective report lacks essential model, data, and validation detail. Multiple images per patient could cross data splits. Augmentation timing is unspecified. There is no external or prospective validation, calibration, uncertainty, patient-level confusion matrix, or fairness analysis. Institutional results lack sample sizes. Qualitative findings lack quotations and coding detail. Time claims conflict. The conclusion overstates what the design can show, and multiple citation-to-claim mismatches materially weaken confidence.
06
How I contributed
I contributed to the development and communication of this interdisciplinary study, supporting research coordination, critical review, interpretation, and manuscript preparation. I also brought a clinical and public-health perspective to the discussion. Because the paper has important reporting limitations, this page clearly separates participation in the work from claims that would require stronger external validation.
07
Explore and cite the paper
Essa I, Memoona, Malhi SD, Malhi DK, Asghar T, Tahir M, Muhammad SK. Artificial intelligence in early cancer detection: a paradigm shift in oncology diagnostic. Journal of Medical & Health Sciences Review. 2025;2(3):4486-4495. doi:10.62019/22204z93.
This is a plain-language research summary, not personal medical advice. Interpret findings in the context of the design and limitations.