Page 55 - ITPS-8-3

INNOSC Theranostics and
Pharmacological Sciences Precision medicine and beyond in oncology

Figure 3. Current summary of diverse applications of artificial intelligence
Abbreviations: CT: Computed tomography; EKGs: Electrocardiograms; FDA: Food and Drug Administration.

automate the process of detecting and labeling tumor the conventional double reading method. Radiologists in
markers in cancers such as HCC and mesothelioma. the AI-supported group had a detection rate that was 17.6%
132
In addition, numerous studies have demonstrated the (95% confidence interval: 5.7, 30.8) higher than those in
141
diagnostic power of deep learning in histopathology. 137-139 the control group. In the diagnosis of prostate cancer,
For example, Ko et al.’s study revealed how AI can be AI can be used to identify clinically significant lesions to
139
142
used to increase efficiency and accuracy in diagnosing allow more targeted biopsy procedures. This enables
gastrointestinal cancers. In the study, pathologists the clinician to focus on specific areas of the prostate,
136
used an AI-based tool instead of human pathologists potentially reducing the risk of under- and overtreatment.
to double-check the screening of gastric and colorectal 5.2. AI in drug development
biopsy specimens. By incorporating AI into their quality
control protocol, they were able to increase the number AI-based tools can also be used to investigate the
of slides reviewed in the same period by 7–10 times. The progression of carcinogenesis and predict the fitness of
accuracy rates of the gastric and colorectal models were potential anticancer targets. For example, AlphaFold 2 uses
93.08% and 95.03%, respectively. These findings align AI to obtain a protein sequence, predict its backbone shape
with broader trends in the field: a meta-analysis of 48 and side-chain conformations, and subsequently generate
AI-assisted diagnostic pathology studies found a mean a model of the overall protein structure. 143 Increasing the
sensitivity of 96.3% (confidence interval: 94.1–97.7) and a accuracy of structure prediction can help researchers better
mean specificity of 93.3% (confidence interval: 90.5–95.4) understand factors, such as ligand binding and molecular
in disease detection across all studies. 140 function, that contribute to drug-target interactions. Other
AI tools attempt to streamline the process of identifying the
Similar principles driving AI innovations in pathology most promising treatments for specific cancers. PINNED
can be applied to developments in radiology. AI has is one such machine learning model that can be used to
demonstrated promise toward aiding the detection and assess potential anticancer therapies and evaluate the
diagnosis of cancerous lesions in radiological imaging. druggability of potential target proteins by assigning scores
A 2025 study compared breast cancer detection rates based on the proteins’ structure, sequence, localization,
between two groups of radiologists: those who used biological function, and network information. In
144
AI-supported double reading and a control group who used addition, Huang et al.’s MADRIGAL is a multimodal AI
145
Volume 8 Issue 3 (2025) 49 doi: 10.36922/ITPS025140018