BMAT Technology & Digital Applications 3 — Questions and Answers
Question 1: Which branch of artificial intelligence trains models by rewarding desired outputs and penalizing undesired ones through environmental interaction?
- Supervised learning
- Unsupervised learning
- Reinforcement learning (Correct answer)
- Transfer learning
Correct answer: Reinforcement learning
Reinforcement learning uses a reward-penalty signal from the environment to iteratively improve an agent's decision-making policy.
Question 2: A hospital digitizes pathology slides for remote review. The resulting files are termed:
- DICOM images
- Whole slide images (WSI) (Correct answer)
- FASTA sequences
- FASTQ reads
Correct answer: Whole slide images (WSI)
Whole slide images (WSI) are high-resolution digital scans of physical pathology slides used in digital pathology workflows.
Question 3: In genomic sequencing, 'read depth' (coverage) refers to:
- The length of a single DNA fragment sequenced
- How many times each base position is sequenced on average (Correct answer)
- The total number of genes in a genome
- The error rate of the sequencing machine
Correct answer: How many times each base position is sequenced on average
Read depth measures how many independent sequencing reads cover each base, with higher depth improving variant-calling accuracy.
Question 4: Which encryption approach allows a cloud server to perform computations on medical data without ever decrypting it?
- AES-256 encryption
- Homomorphic encryption (Correct answer)
- Public-key infrastructure (PKI)
- SSL/TLS
Correct answer: Homomorphic encryption
Homomorphic encryption enables arithmetic operations on ciphertext, producing encrypted results that match those of operations on plaintext.
Question 5: A radiologist's deep learning model achieves 98% accuracy on its training set but only 72% on new patient scans. This is an example of:
- Underfitting
- Overfitting (Correct answer)
- Data augmentation failure
- Calibration error
Correct answer: Overfitting
Overfitting occurs when a model learns training-specific noise rather than generalizable patterns, leading to poor performance on unseen data.
Question 6: Robot-assisted surgery (e.g., the da Vinci system) primarily benefits surgeons by providing:
- Fully autonomous operative decision-making
- Enhanced dexterity and tremor filtration at a small scale (Correct answer)
- Cheaper procedures than open surgery
- Real-time genetic analysis of tissues
Correct answer: Enhanced dexterity and tremor filtration at a small scale
Robotic surgical systems scale down and filter surgeon hand movements, enabling precise manipulation through small incisions.
Question 7: Which metric best describes the proportion of true positives correctly identified by a diagnostic algorithm among all actual positive cases?
- Specificity
- Positive predictive value
- Sensitivity (recall) (Correct answer)
- F1 score
Correct answer: Sensitivity (recall)
Sensitivity (recall) = TP / (TP + FN), measuring how well the algorithm detects all true positive cases.
BMAT Technology & Digital Applications 3
Which branch of artificial intelligence trains models by rewarding desired outputs and penalizing undesired ones through environmental interaction?