Reliable Breast Cancer Diagnosis with Deep Learning: DCGAN-Driven Mammogram Synthesis and Validity Assessment

Algorithm 1

Pseudo-code for similarity assessment and outliers removal.

	Begin
	CosineSimilarity (A, B) {
	RETURN DotProduct (A, B)/(Norm (A)  Norm (B));
	}
	CalculateMeanVector (Vectors) {
	RETURN average of vectors along each dimension;
	}
	CalculateStandardDeviation (Vectors) {
	RETURN standard deviation of vectors along each dimension;
	}
	CalculateCosineSimilarities (ImageVectors) {
	CosineSimilarities ← Empty list;
	FOR each vector in ImageVectors {
	Cosine ← CosineSimilarity (Vector, CalculateMeanVector (All previous vectors in ImageVectors));
	Append Cosine to CosineSimilarities;
	}
	RETURN CosineSimilarities;
	}
	CalculateThresholds (CosineSimilarities) {
	Mean ← calculate mean of CosineSimilarities;
	StdDev ← CalculateStandardDeviation (CosineSimilarities);
	ThresholdHigh ← mean + 3  StdDev;
	ThresholdLow ← mean − 3  StdDev;
	RETURN ThresholdHigh, ThresholdLow;
	}
	FilterImagesByCosineSimilarity (ImageVectors) {
	CosineSimilarities ← CalculateCosineSimilarities (ImageVectors);
	ThresholdHigh, ThresholdLow ← CalculateThresholds (CosineSimilarities);
	FilteredImages ← Empty list;
	FOR each Cosine in CosineSimilarities {
	IF Cosine is within ThresholdHigh and ThresholdLow {
	Append corresponding ImageVector to FilteredImages;
	}
	}
	RETURN FilteredImages;
	}
	End