EARLY DETECTION OF ORAL CANCER USING CBCT WITH THE ASSISTANCE OF ARTIFICIAL INTELLIGENCE

Oral cancer, seen across much of the world, continues to pose serious public health concerns. High rates of illness and death often stem from the discovery that comes late in the disease’s course. Among malignancies of the mouth, “oral squamous cell carcinoma” holds the highest frequency, and better outcomes are tied closely to early diagnosis. Where detection comes sooner, responses to therapy often show more success, and the chance of survival rises. “Cone Beam Computed Tomography”, a complex imaging system, provides layered three-dimensional views of the oral and facial regions. These images help bring subtle tissue changes to light. Aim of the study: The study aimed to probe how reliably the combined AI-CBCT system could reflect the actual tissue conditions confirmed by histopathological findings. Materials and Methods: A total of 400 individuals participated in the study. Age distribution extended from 18 to 60 years. Gender classification remained unsegregated and did not serve as a variable of analytical weight. From the overall group, 200 exhibited clinical indications suggestive of squamous cell carcinoma localized within the mandibular region and were placed in the experimental group. The other 200 did not show signs of the disease. Those were used as controls. CBCT images were captured to assist in identifying potential malignancies within the jaw structures. Initially, each image underwent examination by a skilled Radiologist; findings were noted in sequence, based on visual inspection and diagnostic experience. Following that, the identical images were processed through an AI-based system. The model, structured on a deep conventional neural framework, had been trained earlier using labeled CBCT scans. These included both normal and carcinoma-affected samples, allowing for pattern recognition across a range of anatomical presentations. The histopathological reports were obtained to confirm the cancerous cases (used as a reference for diagnostic certainty). Once all readings were compiled, from both manual and AI interpretations, comparative analysis was carried out. Statistical tests were used to explore the agreement between Radiologist assessments and AI outputs, in comparison to the histopathological results. Results: Statistical analysis suggested that the contrast in cancer detection within the jaws, comparing AI output to Radiologist evaluations, did not reach the threshold for statistical significance at the 0.05 level (McNemar’s Test = 2.77, p-value: 0.096). But AI performance still edged slightly ahead across all evaluated metrics, in particular, CBCT assessments aided by AI reached a higher specificity of 96.97% versus the Radiologist’s specificity of 86.96% and better overall accuracy, 97.50% versus the Radiologist’s accuracy of 92.00%, these differences, unlike the overall detection rate, did show statistical significance (Z-Score = -2.13, p = 0.034; Z-Score = -2.47, p = 0.014). Sensitivity came out nearly the same between the two. AI at 97.76%. Radiologists at 94.66%. The gap, while visible, didn’t reach statistical significance. Z-score landed at 1.33. p-value read 0.185. Results lean slightly in favour of AI-supported assessments. Not a sweeping lead. Not across all metrics. Just a subtle tilt in performance. Conclusion: The system based on artificial intelligence successfully identified early-stage cancers that are often difficult to detect through standard CBCT scans, where details can remain faint or uncertain. What emerges here is a direction that, while not without challenges, suggests real potential. Joining AI with CBCT points toward a shift, perhaps even a leap, in the broader landscape of oral cancer detection.